{
  "url": "https://epoch.ai/data/benchmark_data.zip",
  "documentation": "https://epoch.ai/benchmarks/use-this-data",
  "license": "CC BY 4.0",
  "licenseUrl": "https://creativecommons.org/licenses/by/4.0/",
  "attribution": "Epoch AI, Capabilities & benchmarking. Retrieved from epoch.ai. MEKreflect has selected fields and reformatted the data; no affiliation or endorsement is implied.",
  "retrievedAt": "2026-10-16T15:15:40.854Z",
  "sourceLastModified": null,
  "archiveSHA256": "63a2858b29f782d8a2d39a0d3feb7752da6e38ab764a24d7e626919a689c61d1",
  "models": [
    {
      "name": "Claude Opus 5.5",
      "modelGroup": "Claude Opus 5.5",
      "score": 166.82,
      "confidenceLow": 163.54,
      "confidenceHigh": 170.99,
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-6 Astra",
      "modelGroup": "GPT-6 Astra",
      "score": 166.41,
      "confidenceLow": 163.35,
      "confidenceHigh": 171.23,
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-6.1 Sol",
      "modelGroup": "GPT-6.1 Sol",
      "score": 166.32,
      "confidenceLow": 163.17,
      "confidenceHigh": 170.82,
      "releaseDate": "2026-09-29",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Fable 5.1",
      "modelGroup": "Claude Fable 5.1",
      "score": 164.65,
      "confidenceLow": 161.92,
      "confidenceHigh": 168.48,
      "releaseDate": "2026-09-01",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 4 Argon",
      "modelGroup": "Gemini 4 Argon",
      "score": 164.56,
      "confidenceLow": 161.67,
      "confidenceHigh": 168.7,
      "releaseDate": "2026-09-30",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "Limited access",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Sonnet 5.5",
      "modelGroup": "Claude Sonnet 5.5",
      "score": 164.45,
      "confidenceLow": 161.35,
      "confidenceHigh": 168.26,
      "releaseDate": "2026-09-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Opus 5",
      "modelGroup": "Claude Opus 5",
      "score": 162.6,
      "confidenceLow": 160,
      "confidenceHigh": 165.75,
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-6 Sol",
      "modelGroup": "GPT-6 Sol",
      "score": 162.55,
      "confidenceLow": 160.03,
      "confidenceHigh": 165.84,
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Fable 5",
      "modelGroup": "Claude Fable 5",
      "score": 162.13,
      "confidenceLow": 159.53,
      "confidenceHigh": 166.32,
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.5 Pro",
      "modelGroup": "GPT-5.5 Pro",
      "score": 161.97,
      "confidenceLow": 158.97,
      "confidenceHigh": 165.09,
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.6 Sol",
      "modelGroup": "GPT-5.6 Sol",
      "score": 161.73,
      "confidenceLow": 159.47,
      "confidenceHigh": 165.04,
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.6 Terra",
      "modelGroup": "GPT-5.6 Terra",
      "score": 159.43,
      "confidenceLow": 156.79,
      "confidenceHigh": 162.18,
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.4 Pro",
      "modelGroup": "GPT-5.4 Pro",
      "score": 159,
      "confidenceLow": 156.48,
      "confidenceHigh": 161.69,
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Haiku 5.5",
      "modelGroup": "Claude Haiku 5.5",
      "score": 158.98,
      "confidenceLow": 156.59,
      "confidenceHigh": 162.07,
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.5",
      "modelGroup": "GPT-5.5",
      "score": 158.95,
      "confidenceLow": 156.4,
      "confidenceHigh": 161.92,
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Opus 4.8",
      "modelGroup": "Claude Opus 4.8",
      "score": 158.26,
      "confidenceLow": 156.24,
      "confidenceHigh": 160.79,
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Kimi K3",
      "modelGroup": "Kimi K3",
      "score": 157.42,
      "confidenceLow": 155.1,
      "confidenceHigh": 160.11,
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "accessibility": "Open weights (non-commercial)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 3.7 Flash",
      "modelGroup": "Gemini 3.7 Flash",
      "score": 157.22,
      "confidenceLow": 155.33,
      "confidenceHigh": 159.68,
      "releaseDate": "2026-08-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 3.8 Flash",
      "modelGroup": "Gemini 3.8 Flash",
      "score": 157.2,
      "confidenceLow": 154.97,
      "confidenceHigh": 160.45,
      "releaseDate": "2026-09-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen 3.8 Max",
      "modelGroup": "Qwen 3.8 Max",
      "score": 156.88,
      "confidenceLow": 155.07,
      "confidenceHigh": 159.09,
      "releaseDate": "2026-08-02",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.3 Codex",
      "modelGroup": "GPT-5.3 Codex",
      "score": 156.76,
      "confidenceLow": 152.86,
      "confidenceHigh": 160.49,
      "releaseDate": "2026-02-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Muse Spark 1.3",
      "modelGroup": "Muse Spark 1.3",
      "score": 156.76,
      "confidenceLow": 154.89,
      "confidenceHigh": 158.88,
      "releaseDate": "2026-09-02",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.4",
      "modelGroup": "GPT-5.4",
      "score": 156.72,
      "confidenceLow": 155.04,
      "confidenceHigh": 158.94,
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Sonnet 5",
      "modelGroup": "Claude Sonnet 5",
      "score": 156.36,
      "confidenceLow": 153.97,
      "confidenceHigh": 158.94,
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.6 Luna",
      "modelGroup": "GPT-5.6 Luna",
      "score": 156.33,
      "confidenceLow": 153.86,
      "confidenceHigh": 159.04,
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Opus 4.7",
      "modelGroup": "Claude Opus 4.7",
      "score": 156.32,
      "confidenceLow": 154.58,
      "confidenceHigh": 158.49,
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-6 Luna",
      "modelGroup": "GPT-6 Luna",
      "score": 156.31,
      "confidenceLow": 153.91,
      "confidenceHigh": 158.75,
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Grok 4.6",
      "modelGroup": "Grok 4.6",
      "score": 156.31,
      "confidenceLow": 154.54,
      "confidenceHigh": 158.67,
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.2 Pro",
      "modelGroup": "GPT-5.2 Pro",
      "score": 156.03,
      "confidenceLow": 153.84,
      "confidenceHigh": 159.56,
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GLM-5.3",
      "modelGroup": "GLM-5.3",
      "score": 155.91,
      "confidenceLow": 154.04,
      "confidenceHigh": 158.39,
      "releaseDate": "2026-08-14",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek V4.1 Flash",
      "modelGroup": "DeepSeek V4.1 Flash",
      "score": 155.57,
      "confidenceLow": 153.51,
      "confidenceHigh": 157.55,
      "releaseDate": "2026-09-09",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek V4 Pro 0813",
      "modelGroup": "DeepSeek V4 Pro 0813",
      "score": 155.29,
      "confidenceLow": 153.53,
      "confidenceHigh": 157.63,
      "releaseDate": "2026-08-13",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Opus 4.6",
      "modelGroup": "Claude Opus 4.6",
      "score": 155.27,
      "confidenceLow": 153.59,
      "confidenceHigh": 157.05,
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3.8 Max (0902)",
      "modelGroup": "Qwen3.8 Max (0902)",
      "score": 154.91,
      "confidenceLow": 153.11,
      "confidenceHigh": 156.7,
      "releaseDate": "2026-09-01",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Muse Spark 1.2",
      "modelGroup": "Muse Spark 1.2",
      "score": 154.9,
      "confidenceLow": 152.96,
      "confidenceHigh": 157.03,
      "releaseDate": "2026-08-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 3.1 Pro",
      "modelGroup": "Gemini 3.1 Pro",
      "score": 154.67,
      "confidenceLow": 152.48,
      "confidenceHigh": 157.11,
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek V4 Flash 0731",
      "modelGroup": "DeepSeek V4 Flash 0731",
      "score": 154.49,
      "confidenceLow": 152.12,
      "confidenceHigh": 156.37,
      "releaseDate": "2026-07-31",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 3.5 Flash",
      "modelGroup": "Gemini 3.5 Flash",
      "score": 154.38,
      "confidenceLow": 152.41,
      "confidenceHigh": 156.5,
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Grok 4.7",
      "modelGroup": "Grok 4.7",
      "score": 154.37,
      "confidenceLow": 152.73,
      "confidenceHigh": 156.49,
      "releaseDate": "2026-09-21",
      "organization": "xAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 3.6 Flash",
      "modelGroup": "Gemini 3.6 Flash",
      "score": 154.24,
      "confidenceLow": 152.74,
      "confidenceHigh": 156.1,
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Muse Spark 1.1",
      "modelGroup": "Muse Spark 1.1",
      "score": 154,
      "confidenceLow": 151.66,
      "confidenceHigh": 156.27,
      "releaseDate": "2026-07-09",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Grok 4.5",
      "modelGroup": "Grok 4.5",
      "score": 153.93,
      "confidenceLow": 152.47,
      "confidenceHigh": 155.89,
      "releaseDate": "2026-07-08",
      "organization": "xAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3.7-Max",
      "modelGroup": "Qwen3.7-Max",
      "score": 153.83,
      "confidenceLow": 151.98,
      "confidenceHigh": 156.24,
      "releaseDate": "2026-05-19",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.2",
      "modelGroup": "GPT-5.2",
      "score": 153.46,
      "confidenceLow": 151.64,
      "confidenceHigh": 155.27,
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GLM-5.3-Flash",
      "modelGroup": "GLM-5.3-Flash",
      "score": 153.23,
      "confidenceLow": 150.95,
      "confidenceHigh": 155.4,
      "releaseDate": "2026-08-20",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 3 Pro",
      "modelGroup": "Gemini 3 Pro",
      "score": 152.9,
      "confidenceLow": 151,
      "confidenceHigh": 154.95,
      "releaseDate": "2025-11-18",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Sonnet 4.6",
      "modelGroup": "Claude Sonnet 4.6",
      "score": 152.39,
      "confidenceLow": 150.04,
      "confidenceHigh": 154.65,
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Muse Spark",
      "modelGroup": "Muse Spark",
      "score": 152.13,
      "confidenceLow": 150.33,
      "confidenceHigh": 155.41,
      "releaseDate": "2026-04-08",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Grok 4.20",
      "modelGroup": "Grok 4.20",
      "score": 152.08,
      "confidenceLow": 149.51,
      "confidenceHigh": 154.2,
      "releaseDate": "2026-02-17",
      "organization": "xAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GLM-5.2",
      "modelGroup": "GLM-5.2",
      "score": 151.86,
      "confidenceLow": 150.07,
      "confidenceHigh": 154,
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 3 Flash",
      "modelGroup": "Gemini 3 Flash",
      "score": 151.8,
      "confidenceLow": 150.54,
      "confidenceHigh": 153.39,
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen 3.8 27B",
      "modelGroup": "Qwen 3.8 27B",
      "score": 151.55,
      "confidenceLow": 149.75,
      "confidenceHigh": 153.14,
      "releaseDate": "2026-08-14",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Kimi K2.6",
      "modelGroup": "Kimi K2.6",
      "score": 150.92,
      "confidenceLow": 149.12,
      "confidenceHigh": 152.75,
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5 Pro",
      "modelGroup": "GPT-5 Pro",
      "score": 150.74,
      "confidenceLow": 148.91,
      "confidenceHigh": 154,
      "releaseDate": "2025-10-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Inkling-Small",
      "modelGroup": "Inkling-Small",
      "score": 150.18,
      "confidenceLow": 147.31,
      "confidenceHigh": 152.05,
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Opus 4.5",
      "modelGroup": "Claude Opus 4.5",
      "score": 150.17,
      "confidenceLow": 147.95,
      "confidenceHigh": 152.82,
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5",
      "modelGroup": "GPT-5",
      "score": 150,
      "confidenceLow": null,
      "confidenceHigh": null,
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GLM-5.1",
      "modelGroup": "GLM-5.1",
      "score": 149.99,
      "confidenceLow": 148.05,
      "confidenceHigh": 151.77,
      "releaseDate": "2026-04-07",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Kimi K2.7 Code",
      "modelGroup": "Kimi K2.7 Code",
      "score": 149.87,
      "confidenceLow": 147.94,
      "confidenceHigh": 151.71,
      "releaseDate": "2026-06-12",
      "organization": "Moonshot",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.1",
      "modelGroup": "GPT-5.1",
      "score": 149.76,
      "confidenceLow": 148.47,
      "confidenceHigh": 151.47,
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen 3.6 Max (Preview)",
      "modelGroup": "Qwen 3.6 Max (Preview)",
      "score": 149.24,
      "confidenceLow": 147.49,
      "confidenceHigh": 152.17,
      "releaseDate": "2026-04-20",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-V4-Pro",
      "modelGroup": "DeepSeek-V4-Pro",
      "score": 149.19,
      "confidenceLow": 147.41,
      "confidenceHigh": 150.72,
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Grok 4.3 Beta",
      "modelGroup": "Grok 4.3 Beta",
      "score": 148.95,
      "confidenceLow": 147.27,
      "confidenceHigh": 150.34,
      "releaseDate": "2026-04-17",
      "organization": "xAI",
      "country": "United States of America",
      "accessibility": "Hosted access (no API)",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.4 Mini",
      "modelGroup": "GPT-5.4 Mini",
      "score": 148.77,
      "confidenceLow": 147.08,
      "confidenceHigh": 150.5,
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Inkling",
      "modelGroup": "Inkling",
      "score": 148.59,
      "confidenceLow": 146.08,
      "confidenceHigh": 150.55,
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Kimi K2.5",
      "modelGroup": "Kimi K2.5",
      "score": 147.98,
      "confidenceLow": 146.36,
      "confidenceHigh": 149.3,
      "releaseDate": "2026-01-27",
      "organization": "Moonshot",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen 3.6 Plus",
      "modelGroup": "Qwen 3.6 Plus",
      "score": 147.49,
      "confidenceLow": 145.29,
      "confidenceHigh": 149.27,
      "releaseDate": "2026-03-31",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "o3-pro",
      "modelGroup": "o3-pro",
      "score": 147.37,
      "confidenceLow": 145.7,
      "confidenceHigh": 150.18,
      "releaseDate": "2025-06-10",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3.7-Plus",
      "modelGroup": "Qwen3.7-Plus",
      "score": 147.3,
      "confidenceLow": 145.59,
      "confidenceHigh": 148.83,
      "releaseDate": "2026-06-02",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "MiniMax-M3",
      "modelGroup": "MiniMax-M3",
      "score": 147.13,
      "confidenceLow": 142.79,
      "confidenceHigh": 150,
      "releaseDate": "2026-06-01",
      "organization": "MiniMax",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral Large 4",
      "modelGroup": "Mistral Large 4",
      "score": 146.9,
      "confidenceLow": 143.68,
      "confidenceHigh": 150.94,
      "releaseDate": "2026-10-06",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "o3",
      "modelGroup": "o3",
      "score": 146.87,
      "confidenceLow": 144.95,
      "confidenceHigh": 148.72,
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Sonnet 4.5",
      "modelGroup": "Claude Sonnet 4.5",
      "score": 146.83,
      "confidenceLow": 145.2,
      "confidenceHigh": 148.4,
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "MiniMax-M2.5",
      "modelGroup": "MiniMax-M2.5",
      "score": 146.67,
      "confidenceLow": 142.63,
      "confidenceHigh": 147.94,
      "releaseDate": "2026-02-12",
      "organization": "MiniMax",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3.5 397B-A17B",
      "modelGroup": "Qwen3.5 397B-A17B",
      "score": 146.66,
      "confidenceLow": 144.96,
      "confidenceHigh": 148.32,
      "releaseDate": "2026-02-13",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3.6 27B",
      "modelGroup": "Qwen3.6 27B",
      "score": 146.48,
      "confidenceLow": 144.12,
      "confidenceHigh": 147.82,
      "releaseDate": "2026-04-22",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Grok 4",
      "modelGroup": "Grok 4",
      "score": 146.43,
      "confidenceLow": 144.57,
      "confidenceHigh": 148.15,
      "releaseDate": "2025-07-09",
      "organization": "xAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen 3.5 Plus (hosted 397B-A17B)",
      "modelGroup": "Qwen 3.5 Plus (hosted 397B-A17B)",
      "score": 146.32,
      "confidenceLow": 143.7,
      "confidenceHigh": 147.65,
      "releaseDate": "2026-02-16",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.4 Nano",
      "modelGroup": "GPT-5.4 Nano",
      "score": 146.12,
      "confidenceLow": 143.63,
      "confidenceHigh": 148.51,
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-V3.2",
      "modelGroup": "DeepSeek-V3.2",
      "score": 146.11,
      "confidenceLow": 144.41,
      "confidenceHigh": 147.27,
      "releaseDate": "2025-12-01",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-V4-Flash",
      "modelGroup": "DeepSeek-V4-Flash",
      "score": 146.05,
      "confidenceLow": 143.69,
      "confidenceHigh": 148.14,
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Kimi K2 Thinking",
      "modelGroup": "Kimi K2 Thinking",
      "score": 145.96,
      "confidenceLow": 143.39,
      "confidenceHigh": 147.34,
      "releaseDate": "2025-11-06",
      "organization": "Moonshot",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "MiniMax-M2.7",
      "modelGroup": "MiniMax-M2.7",
      "score": 145.88,
      "confidenceLow": 138.51,
      "confidenceHigh": 148.03,
      "releaseDate": "2026-03-18",
      "organization": "MiniMax",
      "country": "China",
      "accessibility": "Open weights (non-commercial)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GLM-5",
      "modelGroup": "GLM-5",
      "score": 145.78,
      "confidenceLow": 143.79,
      "confidenceHigh": 147.65,
      "releaseDate": "2026-02-11",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Nemotron 3 Ultra",
      "modelGroup": "Nemotron 3 Ultra",
      "score": 145.77,
      "confidenceLow": 143.39,
      "confidenceHigh": 147.55,
      "releaseDate": "2026-06-04",
      "organization": "Nvidia",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "o4-mini",
      "modelGroup": "o4-mini",
      "score": 145.6,
      "confidenceLow": 143.32,
      "confidenceHigh": 147.32,
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5 mini",
      "modelGroup": "GPT-5 mini",
      "score": 145.43,
      "confidenceLow": 143.53,
      "confidenceHigh": 146.9,
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.5 Pro (Jun 2025)",
      "modelGroup": "Gemini 2.5 Pro (Jun 2025)",
      "score": 145.31,
      "confidenceLow": 143.54,
      "confidenceHigh": 146.92,
      "releaseDate": "2025-06-05",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 3.5 Flash-Lite",
      "modelGroup": "Gemini 3.5 Flash-Lite",
      "score": 145.15,
      "confidenceLow": 142.7,
      "confidenceHigh": 146.66,
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-V3.2-Exp",
      "modelGroup": "DeepSeek-V3.2-Exp",
      "score": 144.97,
      "confidenceLow": 142.44,
      "confidenceHigh": 146.98,
      "releaseDate": "2025-09-29",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3.7 Flash",
      "modelGroup": "Qwen3.7 Flash",
      "score": 144.64,
      "confidenceLow": 142.62,
      "confidenceHigh": 147.19,
      "releaseDate": "2026-07-27",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 3.1 Flash-Lite",
      "modelGroup": "Gemini 3.1 Flash-Lite",
      "score": 144.47,
      "confidenceLow": 142.28,
      "confidenceHigh": 146.04,
      "releaseDate": "2026-03-03",
      "organization": "Google",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Grok 4 Fast",
      "modelGroup": "Grok 4 Fast",
      "score": 144.19,
      "confidenceLow": 141.95,
      "confidenceHigh": 146.36,
      "releaseDate": "2025-09-19",
      "organization": "xAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.5 Pro (Mar 2025)",
      "modelGroup": "Gemini 2.5 Pro (Mar 2025)",
      "score": 144.12,
      "confidenceLow": 141.88,
      "confidenceHigh": 146.89,
      "releaseDate": "2025-03-31",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen 3.6 35B-A3B",
      "modelGroup": "Qwen 3.6 35B-A3B",
      "score": 144.11,
      "confidenceLow": 141.48,
      "confidenceHigh": 146.29,
      "releaseDate": "2026-04-14",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Opus 4.1",
      "modelGroup": "Claude Opus 4.1",
      "score": 144.09,
      "confidenceLow": 141.95,
      "confidenceHigh": 146.14,
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3-235B-A22B-Thinking (Jul 2025)",
      "modelGroup": "Qwen3-235B-A22B-Thinking (Jul 2025)",
      "score": 143.81,
      "confidenceLow": 142.08,
      "confidenceHigh": 145.33,
      "releaseDate": "2025-07-25",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GLM-4.7",
      "modelGroup": "GLM-4.7",
      "score": 143.79,
      "confidenceLow": 141.73,
      "confidenceHigh": 146.12,
      "releaseDate": "2025-12-22",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen 3.5 Flash (hosted 35B-A3B)",
      "modelGroup": "Qwen 3.5 Flash (hosted 35B-A3B)",
      "score": 143.77,
      "confidenceLow": 141.51,
      "confidenceHigh": 145.3,
      "releaseDate": "2026-02-25",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen 3.6 Flash",
      "modelGroup": "Qwen 3.6 Flash",
      "score": 143.05,
      "confidenceLow": 140.23,
      "confidenceHigh": 144.48,
      "releaseDate": "2026-04-27",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.5 Flash (Sep 2025)",
      "modelGroup": "Gemini 2.5 Flash (Sep 2025)",
      "score": 143,
      "confidenceLow": 139.46,
      "confidenceHigh": 146.61,
      "releaseDate": "2025-09-25",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemma 4 31B IT",
      "modelGroup": "Gemma 4 31B IT",
      "score": 142.86,
      "confidenceLow": 140.4,
      "confidenceHigh": 145.23,
      "releaseDate": "2026-04-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Opus 4",
      "modelGroup": "Claude Opus 4",
      "score": 142.65,
      "confidenceLow": 140.44,
      "confidenceHigh": 144.14,
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3.5-35B-A3B",
      "modelGroup": "Qwen3.5-35B-A3B",
      "score": 142.48,
      "confidenceLow": 140.1,
      "confidenceHigh": 144.86,
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5.5 Instant",
      "modelGroup": "GPT-5.5 Instant",
      "score": 142.45,
      "confidenceLow": 140.09,
      "confidenceHigh": 145.27,
      "releaseDate": "2026-05-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.5 Pro (May 2025)",
      "modelGroup": "Gemini 2.5 Pro (May 2025)",
      "score": 142.44,
      "confidenceLow": 139.02,
      "confidenceHigh": 145.04,
      "releaseDate": "2025-05-06",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Haiku 4.5",
      "modelGroup": "Claude Haiku 4.5",
      "score": 142.4,
      "confidenceLow": 139.33,
      "confidenceHigh": 144.11,
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3-Max",
      "modelGroup": "Qwen3-Max",
      "score": 142.34,
      "confidenceLow": 139.97,
      "confidenceHigh": 144.37,
      "releaseDate": "2025-09-24",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "o1",
      "modelGroup": "o1",
      "score": 141.84,
      "confidenceLow": 139.51,
      "confidenceHigh": 142.88,
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemma 4 26B A4B",
      "modelGroup": "Gemma 4 26B A4B",
      "score": 141.81,
      "confidenceLow": 138.59,
      "confidenceHigh": 143.44,
      "releaseDate": "2026-04-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Sonnet 4",
      "modelGroup": "Claude Sonnet 4",
      "score": 141.67,
      "confidenceLow": 138.71,
      "confidenceHigh": 143.08,
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.5 Flash (May 2025)",
      "modelGroup": "Gemini 2.5 Flash (May 2025)",
      "score": 141.51,
      "confidenceLow": 139.17,
      "confidenceHigh": 142.86,
      "releaseDate": "2025-05-20",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral Medium 3.5",
      "modelGroup": "Mistral Medium 3.5",
      "score": 141.47,
      "confidenceLow": 138.73,
      "confidenceHigh": 144.02,
      "releaseDate": "2026-04-28",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-R1 (May 2025)",
      "modelGroup": "DeepSeek-R1 (May 2025)",
      "score": 141.27,
      "confidenceLow": 138.65,
      "confidenceHigh": 142.79,
      "releaseDate": "2025-05-28",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude 3.7 Sonnet",
      "modelGroup": "Claude 3.7 Sonnet",
      "score": 141.14,
      "confidenceLow": 138.45,
      "confidenceHigh": 142.83,
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.5 Flash (Jun 2025)",
      "modelGroup": "Gemini 2.5 Flash (Jun 2025)",
      "score": 140.77,
      "confidenceLow": 138.55,
      "confidenceHigh": 142.3,
      "releaseDate": "2025-06-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "o3-mini",
      "modelGroup": "o3-mini",
      "score": 140.33,
      "confidenceLow": 137.23,
      "confidenceHigh": 141.9,
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Grok-3 mini",
      "modelGroup": "Grok-3 mini",
      "score": 140.33,
      "confidenceLow": 137.57,
      "confidenceHigh": 142.21,
      "releaseDate": "2025-06-24",
      "organization": "xAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Kimi K2 (Jul 2025)",
      "modelGroup": "Kimi K2 (Jul 2025)",
      "score": 140.12,
      "confidenceLow": 136.23,
      "confidenceHigh": 142.16,
      "releaseDate": "2025-07-12",
      "organization": "Moonshot",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "gpt-oss-120b",
      "modelGroup": "gpt-oss-120b",
      "score": 140,
      "confidenceLow": 135.89,
      "confidenceHigh": 142.38,
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.5 Flash (Apr 2025)",
      "modelGroup": "Gemini 2.5 Flash (Apr 2025)",
      "score": 139.94,
      "confidenceLow": 135.89,
      "confidenceHigh": 141.9,
      "releaseDate": "2025-04-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-V3.1",
      "modelGroup": "DeepSeek-V3.1",
      "score": 139.91,
      "confidenceLow": 135.75,
      "confidenceHigh": 143.23,
      "releaseDate": "2025-08-21",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3-30B-A3B-Thinking (Jul 2025)",
      "modelGroup": "Qwen3-30B-A3B-Thinking (Jul 2025)",
      "score": 139.61,
      "confidenceLow": 135.63,
      "confidenceHigh": 141.01,
      "releaseDate": "2025-07-30",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3.5-9B",
      "modelGroup": "Qwen3.5-9B",
      "score": 139.43,
      "confidenceLow": 136.09,
      "confidenceHigh": 141.08,
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3-235B-A22B",
      "modelGroup": "Qwen3-235B-A22B",
      "score": 139.35,
      "confidenceLow": 135.17,
      "confidenceHigh": 140.83,
      "releaseDate": "2025-04-28",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-5 nano",
      "modelGroup": "GPT-5 nano",
      "score": 139.33,
      "confidenceLow": 134.79,
      "confidenceHigh": 141.4,
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-R1",
      "modelGroup": "DeepSeek-R1",
      "score": 138.96,
      "confidenceLow": 136.14,
      "confidenceHigh": 140.43,
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3-235B-A22B-Instruct (Jul 2025)",
      "modelGroup": "Qwen3-235B-A22B-Instruct (Jul 2025)",
      "score": 138.92,
      "confidenceLow": 135.11,
      "confidenceHigh": 140.81,
      "releaseDate": "2025-07-25",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3-32B",
      "modelGroup": "Qwen3-32B",
      "score": 138.49,
      "confidenceLow": 134.82,
      "confidenceHigh": 140.44,
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Grok 3",
      "modelGroup": "Grok 3",
      "score": 138.32,
      "confidenceLow": 135.52,
      "confidenceHigh": 139.67,
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3-14B",
      "modelGroup": "Qwen3-14B",
      "score": 138.22,
      "confidenceLow": 133.81,
      "confidenceHigh": 140.04,
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "gpt-oss-20b",
      "modelGroup": "gpt-oss-20b",
      "score": 137.81,
      "confidenceLow": 133.2,
      "confidenceHigh": 139.78,
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "QwQ-32B",
      "modelGroup": "QwQ-32B",
      "score": 137.58,
      "confidenceLow": 132.42,
      "confidenceHigh": 141.91,
      "releaseDate": "2025-03-05",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-R1-Distill-Qwen-32B",
      "modelGroup": "DeepSeek-R1-Distill-Qwen-32B",
      "score": 137.42,
      "confidenceLow": 129.51,
      "confidenceHigh": 138.84,
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3-30B-A3B-Instruct (Jul 2025)",
      "modelGroup": "Qwen3-30B-A3B-Instruct (Jul 2025)",
      "score": 137.42,
      "confidenceLow": 132.24,
      "confidenceHigh": 139.57,
      "releaseDate": "2025-07-29",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4.1",
      "modelGroup": "GPT-4.1",
      "score": 136.77,
      "confidenceLow": 133.71,
      "confidenceHigh": 138.58,
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4.5",
      "modelGroup": "GPT-4.5",
      "score": 136.73,
      "confidenceLow": 133.41,
      "confidenceHigh": 138.66,
      "releaseDate": "2025-02-27",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3-30B-A3B",
      "modelGroup": "Qwen3-30B-A3B",
      "score": 136.19,
      "confidenceLow": 130.65,
      "confidenceHigh": 138.66,
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen3-8B",
      "modelGroup": "Qwen3-8B",
      "score": 136.17,
      "confidenceLow": 130.78,
      "confidenceHigh": 138.25,
      "releaseDate": "2025-04-28",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-V3 (Mar 2025)",
      "modelGroup": "DeepSeek-V3 (Mar 2025)",
      "score": 135.93,
      "confidenceLow": 132.49,
      "confidenceHigh": 137.84,
      "releaseDate": "2025-03-24",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "o1-mini",
      "modelGroup": "o1-mini",
      "score": 135.82,
      "confidenceLow": 132.04,
      "confidenceHigh": 137.32,
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-R1-Distill-Qwen-14B",
      "modelGroup": "DeepSeek-R1-Distill-Qwen-14B",
      "score": 135.44,
      "confidenceLow": 127.65,
      "confidenceHigh": 138.1,
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.0 Flash Thinking (Jan 2025)",
      "modelGroup": "Gemini 2.0 Flash Thinking (Jan 2025)",
      "score": 135.36,
      "confidenceLow": 130.1,
      "confidenceHigh": 138.05,
      "releaseDate": "2025-01-21",
      "organization": "Google DeepMind,Google",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.0 Pro",
      "modelGroup": "Gemini 2.0 Pro",
      "score": 135.05,
      "confidenceLow": 131.16,
      "confidenceHigh": 136.86,
      "releaseDate": "2025-02-05",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "Hosted access (no API)",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4.1 mini",
      "modelGroup": "GPT-4.1 mini",
      "score": 135,
      "confidenceLow": 131.19,
      "confidenceHigh": 136.67,
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "o1-preview",
      "modelGroup": "o1-preview",
      "score": 134.78,
      "confidenceLow": 131.1,
      "confidenceHigh": 138.42,
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.0 Flash (Dec 2024)",
      "modelGroup": "Gemini 2.0 Flash (Dec 2024)",
      "score": 134.69,
      "confidenceLow": 124.32,
      "confidenceHigh": 136.59,
      "releaseDate": "2024-12-11",
      "organization": "Google DeepMind,Google",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.0 Flash (Feb 2025)",
      "modelGroup": "Gemini 2.0 Flash (Feb 2025)",
      "score": 134.68,
      "confidenceLow": 131.22,
      "confidenceHigh": 136.61,
      "releaseDate": "2025-02-05",
      "organization": "Google DeepMind,Google",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral Medium 3",
      "modelGroup": "Mistral Medium 3",
      "score": 134.06,
      "confidenceLow": 129.86,
      "confidenceHigh": 135.47,
      "releaseDate": "2025-05-07",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 2.5 Flash-Lite (Jun 2025)",
      "modelGroup": "Gemini 2.5 Flash-Lite (Jun 2025)",
      "score": 133.95,
      "confidenceLow": 130.34,
      "confidenceHigh": 136.5,
      "releaseDate": "2025-06-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude 3.5 Sonnet (October 2024)",
      "modelGroup": "Claude 3.5 Sonnet (October 2024)",
      "score": 133.53,
      "confidenceLow": 128.77,
      "confidenceHigh": 137.81,
      "releaseDate": "2024-10-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Magistral Small 1.0",
      "modelGroup": "Magistral Small 1.0",
      "score": 133.2,
      "confidenceLow": 129.01,
      "confidenceHigh": 134.44,
      "releaseDate": "2025-06-10",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen2.5-Max",
      "modelGroup": "Qwen2.5-Max",
      "score": 132.53,
      "confidenceLow": 128.14,
      "confidenceHigh": 135.49,
      "releaseDate": "2025-01-25",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-V3",
      "modelGroup": "DeepSeek-V3",
      "score": 132.33,
      "confidenceLow": 127.53,
      "confidenceHigh": 135.86,
      "releaseDate": "2024-12-26",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 4 Maverick",
      "modelGroup": "Llama 4 Maverick",
      "score": 132.2,
      "confidenceLow": 127.72,
      "confidenceHigh": 134.08,
      "releaseDate": "2025-04-06",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral Small 3.2",
      "modelGroup": "Mistral Small 3.2",
      "score": 131.75,
      "confidenceLow": 126.77,
      "confidenceHigh": 133.9,
      "releaseDate": "2025-06-20",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 1.5 Pro (Sept 2024)",
      "modelGroup": "Gemini 1.5 Pro (Sept 2024)",
      "score": 131.73,
      "confidenceLow": 127.09,
      "confidenceHigh": 133.14,
      "releaseDate": "2024-09-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Magistral Small 1.2",
      "modelGroup": "Magistral Small 1.2",
      "score": 131.43,
      "confidenceLow": 125.89,
      "confidenceHigh": 133.44,
      "releaseDate": "2025-09-18",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Grok-2 (Dec 2024)",
      "modelGroup": "Grok-2 (Dec 2024)",
      "score": 130.49,
      "confidenceLow": 125.55,
      "confidenceHigh": 132.31,
      "releaseDate": "2024-12-12",
      "organization": "xAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Phi-4",
      "modelGroup": "Phi-4",
      "score": 130.43,
      "confidenceLow": 124.38,
      "confidenceHigh": 132.09,
      "releaseDate": "2024-12-12",
      "organization": "Microsoft Research",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemma 3 27B",
      "modelGroup": "Gemma 3 27B",
      "score": 130.05,
      "confidenceLow": 124.48,
      "confidenceHigh": 132.21,
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude 3.5 Sonnet",
      "modelGroup": "Claude 3.5 Sonnet",
      "score": 130,
      "confidenceLow": null,
      "confidenceHigh": null,
      "releaseDate": "2024-06-20",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 4 Scout",
      "modelGroup": "Llama 4 Scout",
      "score": 129.65,
      "confidenceLow": 124.73,
      "confidenceHigh": 131.27,
      "releaseDate": "2025-04-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4.1 nano",
      "modelGroup": "GPT-4.1 nano",
      "score": 129.63,
      "confidenceLow": 123.53,
      "confidenceHigh": 132.08,
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 1.5 Flash (Sep 2024)",
      "modelGroup": "Gemini 1.5 Flash (Sep 2024)",
      "score": 129.37,
      "confidenceLow": 123.11,
      "confidenceHigh": 131.33,
      "releaseDate": "2024-09-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4o (May 2024)",
      "modelGroup": "GPT-4o (May 2024)",
      "score": 129.05,
      "confidenceLow": 123.66,
      "confidenceHigh": 131.98,
      "releaseDate": "2024-05-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen2.5-72B",
      "modelGroup": "Qwen2.5-72B",
      "score": 129,
      "confidenceLow": 123.57,
      "confidenceHigh": 130.7,
      "releaseDate": "2024-09-19",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4o (Aug 2024)",
      "modelGroup": "GPT-4o (Aug 2024)",
      "score": 128.84,
      "confidenceLow": 123.86,
      "confidenceHigh": 131.09,
      "releaseDate": "2024-08-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 3.1-405B",
      "modelGroup": "Llama 3.1-405B",
      "score": 128.76,
      "confidenceLow": 123.51,
      "confidenceHigh": 130.67,
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4o (Nov 2024)",
      "modelGroup": "GPT-4o (Nov 2024)",
      "score": 128.73,
      "confidenceLow": 123.93,
      "confidenceHigh": 131.2,
      "releaseDate": "2024-11-20",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen2.5-32B",
      "modelGroup": "Qwen2.5-32B",
      "score": 128.53,
      "confidenceLow": 122.49,
      "confidenceHigh": 130.1,
      "releaseDate": "2024-09-17",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral Large 2 (Nov 2024)",
      "modelGroup": "Mistral Large 2 (Nov 2024)",
      "score": 128.53,
      "confidenceLow": 123.54,
      "confidenceHigh": 130.68,
      "releaseDate": "2024-11-18",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (non-commercial)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral Large 2 (Jul 2024)",
      "modelGroup": "Mistral Large 2 (Jul 2024)",
      "score": 127.55,
      "confidenceLow": 122.26,
      "confidenceHigh": 129.57,
      "releaseDate": "2024-07-24",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (non-commercial)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral Small 3.1",
      "modelGroup": "Mistral Small 3.1",
      "score": 127.49,
      "confidenceLow": 122.13,
      "confidenceHigh": 129.45,
      "releaseDate": "2025-03-17",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 3.3 70B",
      "modelGroup": "Llama 3.3 70B",
      "score": 127.34,
      "confidenceLow": 121.94,
      "confidenceHigh": 129.78,
      "releaseDate": "2024-12-06",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4 Turbo (Apr 2024)",
      "modelGroup": "GPT-4 Turbo (Apr 2024)",
      "score": 127.25,
      "confidenceLow": 121.64,
      "confidenceHigh": 128.84,
      "releaseDate": "2024-04-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude 3.5 Haiku",
      "modelGroup": "Claude 3.5 Haiku",
      "score": 127.17,
      "confidenceLow": 120.12,
      "confidenceHigh": 129.66,
      "releaseDate": "2024-10-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral Small 3",
      "modelGroup": "Mistral Small 3",
      "score": 127.08,
      "confidenceLow": 121.21,
      "confidenceHigh": 129.04,
      "releaseDate": "2025-01-30",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 1.5 Pro (May 2024)",
      "modelGroup": "Gemini 1.5 Pro (May 2024)",
      "score": 126.92,
      "confidenceLow": 121.61,
      "confidenceHigh": 129.29,
      "releaseDate": "2024-05-14",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude 3 Opus",
      "modelGroup": "Claude 3 Opus",
      "score": 126.91,
      "confidenceLow": 121.8,
      "confidenceHigh": 129.76,
      "releaseDate": "2024-02-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4o mini",
      "modelGroup": "GPT-4o mini",
      "score": 126.58,
      "confidenceLow": 119.53,
      "confidenceHigh": 128.63,
      "releaseDate": "2024-07-18",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4 Turbo (Nov 2023)",
      "modelGroup": "GPT-4 Turbo (Nov 2023)",
      "score": 126.46,
      "confidenceLow": 121.08,
      "confidenceHigh": 129.16,
      "releaseDate": "2024-01-25",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 3.1-70B",
      "modelGroup": "Llama 3.1-70B",
      "score": 125.94,
      "confidenceLow": 120.41,
      "confidenceHigh": 128.99,
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4 (Mar 2023)",
      "modelGroup": "GPT-4 (Mar 2023)",
      "score": 125.88,
      "confidenceLow": 119.94,
      "confidenceHigh": 130.57,
      "releaseDate": "2023-03-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 3.2 90B",
      "modelGroup": "Llama 3.2 90B",
      "score": 125.51,
      "confidenceLow": 119.55,
      "confidenceHigh": 127.4,
      "releaseDate": "2024-09-24",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen2-72B",
      "modelGroup": "Qwen2-72B",
      "score": 125.28,
      "confidenceLow": 119.02,
      "confidenceHigh": 127.08,
      "releaseDate": "2024-06-07",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-V2 (MoE-236B, May 2024)",
      "modelGroup": "DeepSeek-V2 (MoE-236B, May 2024)",
      "score": 124.78,
      "confidenceLow": 118.34,
      "confidenceHigh": 127.72,
      "releaseDate": "2024-05-07",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Amazon Nova Pro",
      "modelGroup": "Amazon Nova Pro",
      "score": 123.8,
      "confidenceLow": 109.54,
      "confidenceHigh": 126.96,
      "releaseDate": "2024-12-03",
      "organization": "Amazon",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemma 3 12B",
      "modelGroup": "Gemma 3 12B",
      "score": 123.49,
      "confidenceLow": 115.45,
      "confidenceHigh": 128.72,
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-4 (Jun 2023)",
      "modelGroup": "GPT-4 (Jun 2023)",
      "score": 123.16,
      "confidenceLow": 116.28,
      "confidenceHigh": 126.88,
      "releaseDate": "2023-06-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 3-70B",
      "modelGroup": "Llama 3-70B",
      "score": 122.94,
      "confidenceLow": 117.25,
      "confidenceHigh": 126.08,
      "releaseDate": "2024-04-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 1.5 Flash (May 2024)",
      "modelGroup": "Gemini 1.5 Flash (May 2024)",
      "score": 122.61,
      "confidenceLow": 116.7,
      "confidenceHigh": 124.63,
      "releaseDate": "2024-05-23",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemma 2 27B",
      "modelGroup": "Gemma 2 27B",
      "score": 122.1,
      "confidenceLow": 115.12,
      "confidenceHigh": 124.38,
      "releaseDate": "2024-06-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mixtral 8x22B",
      "modelGroup": "Mixtral 8x22B",
      "score": 122.05,
      "confidenceLow": 115.19,
      "confidenceHigh": 124.59,
      "releaseDate": "2024-04-17",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral Large",
      "modelGroup": "Mistral Large",
      "score": 122.03,
      "confidenceLow": 114.45,
      "confidenceHigh": 125.04,
      "releaseDate": "2024-02-26",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "phi-3-small 7.4B",
      "modelGroup": "phi-3-small 7.4B",
      "score": 121.85,
      "confidenceLow": 113.91,
      "confidenceHigh": 124.71,
      "releaseDate": "2024-04-23",
      "organization": "Microsoft",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "phi-3-medium 14B",
      "modelGroup": "phi-3-medium 14B",
      "score": 121.25,
      "confidenceLow": 114.39,
      "confidenceHigh": 124.38,
      "releaseDate": "2024-04-23",
      "organization": "Microsoft",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude 3 Sonnet",
      "modelGroup": "Claude 3 Sonnet",
      "score": 120.72,
      "confidenceLow": 112.47,
      "confidenceHigh": 123.85,
      "releaseDate": "2024-02-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude Instant",
      "modelGroup": "Claude Instant",
      "score": 120.31,
      "confidenceLow": 112.78,
      "confidenceHigh": 123.18,
      "releaseDate": "2023-08-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude 2",
      "modelGroup": "Claude 2",
      "score": 120.15,
      "confidenceLow": 113.73,
      "confidenceHigh": 123.85,
      "releaseDate": "2023-07-11",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemma 2 9B",
      "modelGroup": "Gemma 2 9B",
      "score": 119.85,
      "confidenceLow": 111.52,
      "confidenceHigh": 122.48,
      "releaseDate": "2024-06-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen2.5-Coder-32B",
      "modelGroup": "Qwen2.5-Coder-32B",
      "score": 119.52,
      "confidenceLow": 110.57,
      "confidenceHigh": 124.65,
      "releaseDate": "2024-09-18",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Command R+",
      "modelGroup": "Command R+",
      "score": 119.37,
      "confidenceLow": 108.52,
      "confidenceHigh": 125.39,
      "releaseDate": "2024-08-30",
      "organization": "Cohere,Cohere Labs (formerly Cohere for AI)",
      "country": "Canada",
      "accessibility": "Open weights (non-commercial)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude 2.1",
      "modelGroup": "Claude 2.1",
      "score": 119.3,
      "confidenceLow": 112.16,
      "confidenceHigh": 121.9,
      "releaseDate": "2023-11-21",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral NeMo",
      "modelGroup": "Mistral NeMo",
      "score": 118.71,
      "confidenceLow": 111.04,
      "confidenceHigh": 121.66,
      "releaseDate": "2024-07-18",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-3.5 Turbo (Nov 2023)",
      "modelGroup": "GPT-3.5 Turbo (Nov 2023)",
      "score": 118.58,
      "confidenceLow": 109.72,
      "confidenceHigh": 121.79,
      "releaseDate": "2023-11-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen2.5-7B",
      "modelGroup": "Qwen2.5-7B",
      "score": 118.53,
      "confidenceLow": 110.6,
      "confidenceHigh": 121.29,
      "releaseDate": "2024-09-19",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mixtral 8x7B",
      "modelGroup": "Mixtral 8x7B",
      "score": 118.5,
      "confidenceLow": 110.87,
      "confidenceHigh": 121.19,
      "releaseDate": "2023-12-11",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Claude 3 Haiku",
      "modelGroup": "Claude 3 Haiku",
      "score": 118.39,
      "confidenceLow": 110.36,
      "confidenceHigh": 121.69,
      "releaseDate": "2024-03-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Ministral 3B",
      "modelGroup": "Ministral 3B",
      "score": 118.12,
      "confidenceLow": 105.45,
      "confidenceHigh": 121.97,
      "releaseDate": "2024-10-16",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Yi-34B",
      "modelGroup": "Yi-34B",
      "score": 117.43,
      "confidenceLow": 109.75,
      "confidenceHigh": 120.67,
      "releaseDate": "2023-11-02",
      "organization": "01.AI",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "phi-3-mini 3.8B",
      "modelGroup": "phi-3-mini 3.8B",
      "score": 117.4,
      "confidenceLow": 108.76,
      "confidenceHigh": 120.7,
      "releaseDate": "2024-04-23",
      "organization": "Microsoft",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Stable Beluga 2",
      "modelGroup": "Stable Beluga 2",
      "score": 117.16,
      "confidenceLow": 109.2,
      "confidenceHigh": 120.16,
      "releaseDate": "2023-07-20",
      "organization": "Stability AI",
      "country": "United Kingdom",
      "accessibility": "Open weights (non-commercial)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemini 1.0 Pro",
      "modelGroup": "Gemini 1.0 Pro",
      "score": 117.07,
      "confidenceLow": 109.8,
      "confidenceHigh": 119.8,
      "releaseDate": "2023-12-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 3.1-8B",
      "modelGroup": "Llama 3.1-8B",
      "score": 116.61,
      "confidenceLow": 106.33,
      "confidenceHigh": 121.57,
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 3-8B",
      "modelGroup": "Llama 3-8B",
      "score": 116.49,
      "confidenceLow": 107.81,
      "confidenceHigh": 119.39,
      "releaseDate": "2024-04-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen2.5-Coder-14B",
      "modelGroup": "Qwen2.5-Coder-14B",
      "score": 116.44,
      "confidenceLow": 106.24,
      "confidenceHigh": 122.35,
      "releaseDate": "2024-09-18",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemma 3 4B",
      "modelGroup": "Gemma 3 4B",
      "score": 116.01,
      "confidenceLow": 97.97,
      "confidenceHigh": 123.34,
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-3.5 Turbo (Jan 2024)",
      "modelGroup": "GPT-3.5 Turbo (Jan 2024)",
      "score": 115.74,
      "confidenceLow": 108.01,
      "confidenceHigh": 119.43,
      "releaseDate": "2024-01-25",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "PaLM 2-L",
      "modelGroup": "PaLM 2-L",
      "score": 115.07,
      "confidenceLow": 106.61,
      "confidenceHigh": 127.9,
      "releaseDate": "2023-05-17",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 2-70B",
      "modelGroup": "Llama 2-70B",
      "score": 113.84,
      "confidenceLow": 105.57,
      "confidenceHigh": 117.97,
      "releaseDate": "2023-07-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "GPT-3.5 Turbo (Jun 2023)",
      "modelGroup": "GPT-3.5 Turbo (Jun 2023)",
      "score": 113.42,
      "confidenceLow": 104.8,
      "confidenceHigh": 118.03,
      "releaseDate": "2023-06-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "accessibility": "API access",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen2.5-Coder (7B)",
      "modelGroup": "Qwen2.5-Coder (7B)",
      "score": 113.2,
      "confidenceLow": 102.14,
      "confidenceHigh": 119.29,
      "releaseDate": "2024-09-18",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen-14B",
      "modelGroup": "Qwen-14B",
      "score": 113.09,
      "confidenceLow": 104.56,
      "confidenceHigh": 117.65,
      "releaseDate": "2023-09-24",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral 7B v0.1",
      "modelGroup": "Mistral 7B v0.1",
      "score": 112.27,
      "confidenceLow": 103.3,
      "confidenceHigh": 116.49,
      "releaseDate": "2023-09-27",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Falcon-180B",
      "modelGroup": "Falcon-180B",
      "score": 112.19,
      "confidenceLow": 102.42,
      "confidenceHigh": 119.32,
      "releaseDate": "2023-09-06",
      "organization": "Technology Innovation Institute",
      "country": "United Arab Emirates",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "internlm-20b",
      "modelGroup": "internlm-20b",
      "score": 112.15,
      "confidenceLow": 103.19,
      "confidenceHigh": 117.04,
      "releaseDate": "2023-09-18",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemma 7B",
      "modelGroup": "Gemma 7B",
      "score": 112.05,
      "confidenceLow": 102.44,
      "confidenceHigh": 116.38,
      "releaseDate": "2024-02-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek LLM 67B",
      "modelGroup": "DeepSeek LLM 67B",
      "score": 110.47,
      "confidenceLow": 94.99,
      "confidenceHigh": 117.25,
      "releaseDate": "2023-11-29",
      "organization": "DeepSeek",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "LLaMA-65B",
      "modelGroup": "LLaMA-65B",
      "score": 110.23,
      "confidenceLow": 101.57,
      "confidenceHigh": 114.9,
      "releaseDate": "2023-02-24",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (non-commercial)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Falcon 2 11B",
      "modelGroup": "Falcon 2 11B",
      "score": 109.63,
      "confidenceLow": 100.92,
      "confidenceHigh": 114.9,
      "releaseDate": "2024-05-09",
      "organization": "Technology Innovation Institute",
      "country": "United Arab Emirates",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Mistral 7B v0.3",
      "modelGroup": "Mistral 7B v0.3",
      "score": 109.06,
      "confidenceLow": 97.34,
      "confidenceHigh": 112.81,
      "releaseDate": "2024-05-27",
      "organization": "Mistral AI",
      "country": "France",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek-Coder-V2-Lite-Base",
      "modelGroup": "DeepSeek-Coder-V2-Lite-Base",
      "score": 108.96,
      "confidenceLow": 98.93,
      "confidenceHigh": 114.32,
      "releaseDate": "2024-06-13",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "PaLM 2-M",
      "modelGroup": "PaLM 2-M",
      "score": 108.37,
      "confidenceLow": 99.17,
      "confidenceHigh": 117.72,
      "releaseDate": "2023-05-17",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Phi-2",
      "modelGroup": "Phi-2",
      "score": 108.02,
      "confidenceLow": 85.14,
      "confidenceHigh": 114.08,
      "releaseDate": "2023-12-12",
      "organization": "Microsoft",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen2.5-Coder-3B",
      "modelGroup": "Qwen2.5-Coder-3B",
      "score": 107.82,
      "confidenceLow": 93.77,
      "confidenceHigh": 115.91,
      "releaseDate": "2024-09-18",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Nemotron-4 15B",
      "modelGroup": "Nemotron-4 15B",
      "score": 107.8,
      "confidenceLow": 98.61,
      "confidenceHigh": 112.99,
      "releaseDate": "2024-02-26",
      "organization": "Nvidia",
      "country": "United States of America",
      "accessibility": "Unreleased",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Yi-9B",
      "modelGroup": "Yi-9B",
      "score": 107.69,
      "confidenceLow": 98.08,
      "confidenceHigh": 115.67,
      "releaseDate": "2024-03-01",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "LLaMA-33B",
      "modelGroup": "LLaMA-33B",
      "score": 107.49,
      "confidenceLow": 98.33,
      "confidenceHigh": 112.27,
      "releaseDate": "2023-02-24",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (non-commercial)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen-7B",
      "modelGroup": "Qwen-7B",
      "score": 106.91,
      "confidenceLow": 96.06,
      "confidenceHigh": 112.26,
      "releaseDate": "2023-09-28",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "PaLM 2-S",
      "modelGroup": "PaLM 2-S",
      "score": 106.26,
      "confidenceLow": 96.62,
      "confidenceHigh": 112.9,
      "releaseDate": "2023-05-17",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 2-13B",
      "modelGroup": "Llama 2-13B",
      "score": 106.25,
      "confidenceLow": 96.69,
      "confidenceHigh": 111.49,
      "releaseDate": "2023-07-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 2-34B",
      "modelGroup": "Llama 2-34B",
      "score": 105.29,
      "confidenceLow": 95.69,
      "confidenceHigh": 110.57,
      "releaseDate": "2023-07-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Unreleased",
      "accessGroup": "Closed weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "StarCoder 2 15B",
      "modelGroup": "StarCoder 2 15B",
      "score": 105.09,
      "confidenceLow": 94.1,
      "confidenceHigh": 112.2,
      "releaseDate": "2024-02-20",
      "organization": "Hugging Face,ServiceNow,Nvidia,BigCode",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Yi 6B",
      "modelGroup": "Yi 6B",
      "score": 104.86,
      "confidenceLow": 94.63,
      "confidenceHigh": 110.37,
      "releaseDate": "2023-11-22",
      "organization": "01.AI",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Falcon-40B",
      "modelGroup": "Falcon-40B",
      "score": 104.54,
      "confidenceLow": 94.97,
      "confidenceHigh": 110.17,
      "releaseDate": "2023-05-25",
      "organization": "Technology Innovation Institute",
      "country": "United Arab Emirates",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Baichuan2-13B",
      "modelGroup": "Baichuan2-13B",
      "score": 103.29,
      "confidenceLow": 92.67,
      "confidenceHigh": 110.61,
      "releaseDate": "2023-09-06",
      "organization": "Baichuan",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen2.5-Coder (1.5B)",
      "modelGroup": "Qwen2.5-Coder (1.5B)",
      "score": 103.09,
      "confidenceLow": 88.02,
      "confidenceHigh": 111.76,
      "releaseDate": "2024-09-18",
      "organization": "Alibaba",
      "country": "China",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "internlm-7b",
      "modelGroup": "internlm-7b",
      "score": 102.97,
      "confidenceLow": 92.13,
      "confidenceHigh": 109.55,
      "releaseDate": "2023-07-05",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 3.2 1B",
      "modelGroup": "Llama 3.2 1B",
      "score": 102.04,
      "confidenceLow": 90.32,
      "confidenceHigh": 115.01,
      "releaseDate": "2024-09-24",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "INTELLECT-1",
      "modelGroup": "INTELLECT-1",
      "score": 100.97,
      "confidenceLow": 90.22,
      "confidenceHigh": 106.46,
      "releaseDate": "2024-11-29",
      "organization": "Prime Intellect,Hugging Face,Arcee AI",
      "country": "United States of America",
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "MPT-30B",
      "modelGroup": "MPT-30B",
      "score": 100.76,
      "confidenceLow": 90.55,
      "confidenceHigh": 106.4,
      "releaseDate": "2023-06-22",
      "organization": "MosaicML",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "LLaMA-13B",
      "modelGroup": "LLaMA-13B",
      "score": 100.69,
      "confidenceLow": 90.04,
      "confidenceHigh": 106.64,
      "releaseDate": "2023-02-24",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (non-commercial)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Llama 2-7B",
      "modelGroup": "Llama 2-7B",
      "score": 99.18,
      "confidenceLow": 88.06,
      "confidenceHigh": 105.33,
      "releaseDate": "2023-07-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "chatglm2-6b",
      "modelGroup": "chatglm2-6b",
      "score": 99.05,
      "confidenceLow": 81.48,
      "confidenceHigh": 106.4,
      "releaseDate": "2023-06-24",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "LLaMA-7B",
      "modelGroup": "LLaMA-7B",
      "score": 96.82,
      "confidenceLow": 84.07,
      "confidenceHigh": 103.99,
      "releaseDate": "2023-02-24",
      "organization": "Meta AI",
      "country": "United States of America",
      "accessibility": "Open weights (non-commercial)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Baichuan 2-7B",
      "modelGroup": "Baichuan 2-7B",
      "score": 96.45,
      "confidenceLow": 84.31,
      "confidenceHigh": 104.76,
      "releaseDate": "2023-09-20",
      "organization": "Baichuan",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek Coder 33B",
      "modelGroup": "DeepSeek Coder 33B",
      "score": 96.45,
      "confidenceLow": 84.2,
      "confidenceHigh": 103.38,
      "releaseDate": "2023-11-02",
      "organization": "DeepSeek,Peking University",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Falcon-7B",
      "modelGroup": "Falcon-7B",
      "score": 95.23,
      "confidenceLow": 82.44,
      "confidenceHigh": 102.38,
      "releaseDate": "2023-04-24",
      "organization": "Technology Innovation Institute",
      "country": "United Arab Emirates",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "CodeQwen1.5-7B",
      "modelGroup": "CodeQwen1.5-7B",
      "score": 94.99,
      "confidenceLow": 80.78,
      "confidenceHigh": 103.36,
      "releaseDate": "2024-04-15",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "vicuna-13b-v1.1",
      "modelGroup": "vicuna-13b-v1.1",
      "score": 94.8,
      "confidenceLow": 80.99,
      "confidenceHigh": 102.84,
      "releaseDate": "2023-04-12",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "MPT-7B",
      "modelGroup": "MPT-7B",
      "score": 94.73,
      "confidenceLow": 81.66,
      "confidenceHigh": 102.66,
      "releaseDate": "2023-05-05",
      "organization": "MosaicML",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Gemma 2B",
      "modelGroup": "Gemma 2B",
      "score": 94.34,
      "confidenceLow": 82.8,
      "confidenceHigh": 101.18,
      "releaseDate": "2024-02-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "StarCoder 2 7B",
      "modelGroup": "StarCoder 2 7B",
      "score": 93.7,
      "confidenceLow": 79.05,
      "confidenceHigh": 102.01,
      "releaseDate": "2024-02-20",
      "organization": "Hugging Face,ServiceNow,Nvidia,BigCode",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "XGen-7B",
      "modelGroup": "XGen-7B",
      "score": 93.54,
      "confidenceLow": 81.11,
      "confidenceHigh": 101.04,
      "releaseDate": "2023-06-27",
      "organization": "Salesforce",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen-1_8B",
      "modelGroup": "Qwen-1_8B",
      "score": 93.07,
      "confidenceLow": 71.44,
      "confidenceHigh": 102,
      "releaseDate": "2023-11-30",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "open_llama_7b",
      "modelGroup": "open_llama_7b",
      "score": 91.8,
      "confidenceLow": 77.35,
      "confidenceHigh": 99.47,
      "releaseDate": "2023-06-07",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Phi-1.5",
      "modelGroup": "Phi-1.5",
      "score": 91.68,
      "confidenceLow": 71.13,
      "confidenceHigh": 101.34,
      "releaseDate": "2023-09-11",
      "organization": "Microsoft",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Baichuan1-7B",
      "modelGroup": "Baichuan1-7B",
      "score": 90.61,
      "confidenceLow": 77.57,
      "confidenceHigh": 99.38,
      "releaseDate": "2023-06-01",
      "organization": "Baichuan",
      "country": "China",
      "accessibility": "Open weights (non-commercial)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "RedPajama-INCITE-7B-Base",
      "modelGroup": "RedPajama-INCITE-7B-Base",
      "score": 90.58,
      "confidenceLow": 76.49,
      "confidenceHigh": 99.17,
      "releaseDate": "2023-05-04",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Dolly 2.0-12b",
      "modelGroup": "Dolly 2.0-12b",
      "score": 89.83,
      "confidenceLow": 76.06,
      "confidenceHigh": 97.97,
      "releaseDate": "2023-04-11",
      "organization": "Databricks",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek Coder 6.7B",
      "modelGroup": "DeepSeek Coder 6.7B",
      "score": 89.77,
      "confidenceLow": 75.59,
      "confidenceHigh": 98.06,
      "releaseDate": "2023-11-02",
      "organization": "DeepSeek,Peking University",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "StarCoder 2 3B",
      "modelGroup": "StarCoder 2 3B",
      "score": 88.92,
      "confidenceLow": 73.32,
      "confidenceHigh": 97.85,
      "releaseDate": "2024-02-22",
      "organization": "Hugging Face,ServiceNow,Nvidia,BigCode",
      "country": "United States of America",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Qwen2.5-Coder-0.5B",
      "modelGroup": "Qwen2.5-Coder-0.5B",
      "score": 88.31,
      "confidenceLow": 63.95,
      "confidenceHigh": 100.06,
      "releaseDate": "2024-09-18",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "Cerebras-GPT-13B",
      "modelGroup": "Cerebras-GPT-13B",
      "score": 83.42,
      "confidenceLow": 67.04,
      "confidenceHigh": 92.07,
      "releaseDate": "2023-03-20",
      "organization": "Cerebras Systems",
      "country": "United States of America",
      "accessibility": "Open weights (unrestricted)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "DeepSeek Coder 1.3B",
      "modelGroup": "DeepSeek Coder 1.3B",
      "score": 63.88,
      "confidenceLow": 45.1,
      "confidenceHigh": 76.94,
      "releaseDate": "2023-11-02",
      "organization": "DeepSeek,Peking University",
      "country": "China",
      "accessibility": "Open weights (restricted use)",
      "accessGroup": "Open weights",
      "sourceUrl": "https://epoch.ai/eci"
    },
    {
      "name": "stablelm-tuned-alpha-7b",
      "modelGroup": "stablelm-tuned-alpha-7b",
      "score": 56.05,
      "confidenceLow": 31.31,
      "confidenceHigh": 73.19,
      "releaseDate": "2023-04-19",
      "organization": null,
      "country": null,
      "accessibility": "Unknown",
      "accessGroup": "Other",
      "sourceUrl": "https://epoch.ai/eci"
    }
  ],
  "benchmarks": [
    {
      "name": "GPQA diamond",
      "sourceFile": "gpqa_diamond.csv",
      "scale": 1,
      "randomBaseline": 0.25,
      "scoreCeiling": 1,
      "releaseDate": "2023-11-20",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "FrontierMath-2025-02-28-Private",
      "sourceFile": "frontiermath.csv",
      "scale": 1,
      "randomBaseline": 0,
      "scoreCeiling": 0.57,
      "releaseDate": "2025-02-28",
      "supersededBy": "FrontierMath-Tiers-1-3-v2-Private",
      "inECI": true
    },
    {
      "name": "FrontierMath-Tier-4-2025-07-01-Private",
      "sourceFile": "frontiermath_tier_4.csv",
      "scale": 1,
      "randomBaseline": 0,
      "scoreCeiling": 0.6,
      "releaseDate": "2025-07-11",
      "supersededBy": "FrontierMath-Tier-4-v2-Private",
      "inECI": true
    },
    {
      "name": "FrontierMath-Tiers-1-3-v2-Private",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv",
      "scale": 1,
      "randomBaseline": 0,
      "scoreCeiling": 1,
      "releaseDate": "2026-06-12",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "FrontierMath-Tier-4-v2-Private",
      "sourceFile": "frontiermath_tier_4_v2.csv",
      "scale": 1,
      "randomBaseline": 0,
      "scoreCeiling": 1,
      "releaseDate": "2026-06-12",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "MATH level 5",
      "sourceFile": "math_level_5.csv",
      "scale": 1,
      "randomBaseline": 0,
      "scoreCeiling": 1,
      "releaseDate": "2021-03-05",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "OTIS Mock AIME 2024-2025",
      "sourceFile": "otis_mock_aime_2024_2025.csv",
      "scale": 1,
      "randomBaseline": 0.001,
      "scoreCeiling": 1,
      "releaseDate": "2024-12-19",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "SWE-Bench verified",
      "sourceFile": "swe_bench_verified.csv",
      "scale": 1,
      "randomBaseline": 0,
      "scoreCeiling": 1,
      "releaseDate": "2024-08-13",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "SimpleQA Verified",
      "sourceFile": "simpleqa_verified.csv",
      "scale": 1,
      "randomBaseline": 0,
      "scoreCeiling": 1,
      "releaseDate": "2025-09-09",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "Chess Puzzles",
      "sourceFile": "chess_puzzles.csv",
      "scale": 1,
      "randomBaseline": 0.0496,
      "scoreCeiling": 1,
      "releaseDate": "2025-12-10",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "EBR-bench",
      "sourceFile": "ebr_bench.csv",
      "scale": 1,
      "randomBaseline": 0,
      "scoreCeiling": 1,
      "releaseDate": "2026-07-01",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "Mystery Game Puzzles",
      "sourceFile": "mystery_game_puzzles.csv",
      "scale": 1,
      "randomBaseline": 0.0922,
      "scoreCeiling": 1,
      "releaseDate": "2026-07-30",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "MirrorCode",
      "sourceFile": "mirrorcode.csv",
      "scale": 1,
      "randomBaseline": 0,
      "scoreCeiling": 1,
      "releaseDate": "2026-06-26",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "Furniture Assembly",
      "sourceFile": "furniture_assembly.csv",
      "scale": 1,
      "randomBaseline": 0.3,
      "scoreCeiling": 1,
      "releaseDate": "2026-09-23",
      "supersededBy": null,
      "inECI": true
    },
    {
      "name": "SpireClimb",
      "sourceFile": "spireclimb.csv",
      "scale": 1,
      "randomBaseline": 0,
      "scoreCeiling": 1,
      "releaseDate": "2026-10-14",
      "supersededBy": null,
      "inECI": true
    }
  ],
  "runs": [
    {
      "benchmark": "GPQA diamond",
      "model": "Muse-Spark-Open-816B-A42B_xhigh",
      "score": 0.8857323232323232,
      "bestScore": 0.8857323232323232,
      "standardError": 0.018706209910489154,
      "startedAt": "2026-10-15T18:52:04.000Z",
      "releaseDate": "2026-10-15",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "JnKpor8c3EKWGvmyjLFVzH",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-4-argon_high",
      "score": 0.9640151515151515,
      "bestScore": 0.9640151515151515,
      "standardError": 0.011918129589602694,
      "startedAt": "2026-10-13T17:15:08.000Z",
      "releaseDate": "2026-09-30",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "QEouTvczNsxgkR4i7i8wBF",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "mistral-large-4-0_high",
      "score": 0.8806818181818182,
      "bestScore": 0.8806818181818182,
      "standardError": 0.019328132475292338,
      "startedAt": "2026-10-12T21:38:22.000Z",
      "releaseDate": "2026-10-06",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "Ekx5phukyiJhAcQodMfS2p",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "deepseek-v4.1-flash_max",
      "score": 0.8983585858585859,
      "bestScore": 0.8983585858585859,
      "standardError": 0.017287759372442393,
      "startedAt": "2026-10-09T14:38:01.000Z",
      "releaseDate": "2026-09-09",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "ZU5ebpDaDPiDZLgv7cG8B9",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-haiku-5-5_max",
      "score": 0.8964646464646465,
      "bestScore": 0.8964646464646465,
      "standardError": 0.019774484382483127,
      "startedAt": "2026-10-09T14:35:53.000Z",
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZEKWTGJiWFscHimMkiUmec",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-6.1-sol_max",
      "score": 0.9539141414141414,
      "bestScore": 0.9539141414141414,
      "standardError": 0.013780748450364692,
      "startedAt": "2026-09-29T22:00:34.000Z",
      "releaseDate": "2026-09-29",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LspviLnbeZ82WCngWp6rwR",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-5-5_max",
      "score": 0.9558080808080808,
      "bestScore": 0.9558080808080808,
      "standardError": 0.013710320714521202,
      "startedAt": "2026-09-29T15:15:14.000Z",
      "releaseDate": "2026-09-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "QPk8jbJvo4986sbWtdte6J",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-6-luna_max",
      "score": 0.9046717171717171,
      "bestScore": 0.9046717171717171,
      "standardError": 0.018537378631915503,
      "startedAt": "2026-09-22T22:14:05.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "7F3QFMAkwnCV62HcvxP96q",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "grok-4.7_xhigh",
      "score": 0.92739898989899,
      "bestScore": 0.92739898989899,
      "standardError": 0.015505824295534454,
      "startedAt": "2026-09-22T21:49:22.000Z",
      "releaseDate": "2026-09-21",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hxAhYJfXdPNanxatXpG5hC",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-6-sol_max",
      "score": 0.9431818181818182,
      "bestScore": 0.9431818181818182,
      "standardError": 0.014938512008159717,
      "startedAt": "2026-09-22T19:59:24.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "aJD7ZfFftu2cZoToJhhzF7",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-5-5_max",
      "score": 0.9059343434343434,
      "bestScore": 0.9059343434343434,
      "standardError": 0.019434397190192135,
      "startedAt": "2026-09-22T18:27:39.000Z",
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6gmbahPSxHf9xVhAYqpnhi",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.8-max-0902_xhigh",
      "score": 0.922979797979798,
      "bestScore": 0.922979797979798,
      "standardError": 0.017249976996605708,
      "startedAt": "2026-09-02T18:04:58.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "VzeQXKBH9maR2nDcZJ8b7S",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.8-flash_high",
      "score": 0.9539141414141414,
      "bestScore": 0.9539141414141414,
      "standardError": 0.01401136318210652,
      "startedAt": "2026-09-02T17:15:36.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Br3mjs2GrYcCwX9AMwcihU",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "granite-4.0-micro",
      "score": 0.2828282828282828,
      "bestScore": 0.2828282828282828,
      "standardError": 0.018237830822136196,
      "startedAt": "2026-08-30T22:46:42.000Z",
      "releaseDate": "2025-10-02",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSadaSwnSdMpYRstc8zvgjJ.eval",
      "sourceId": "SadaSwnSdMpYRstc8zvgjJ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "seed-oss-36b-instruct",
      "score": 0.7152777777777778,
      "bestScore": 0.7152777777777778,
      "standardError": 0.02741123270979468,
      "startedAt": "2026-08-30T22:46:22.000Z",
      "releaseDate": "2025-08-20",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FBVat526nF6NUGoVaLCp7DT.eval",
      "sourceId": "BVat526nF6NUGoVaLCp7DT",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Qwen3-32B_none",
      "score": 0.5410353535353535,
      "bestScore": 0.5410353535353535,
      "standardError": 0.028146234201602498,
      "startedAt": "2026-08-30T22:45:08.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FEXpHB6kFqDjRfG9A5LsDf6.eval",
      "sourceId": "EXpHB6kFqDjRfG9A5LsDf6",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-30b-a3b-thinking-2507",
      "score": 0.7007575757575758,
      "bestScore": 0.7007575757575758,
      "standardError": 0.028327897613545493,
      "startedAt": "2026-08-30T22:44:24.000Z",
      "releaseDate": "2025-07-30",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2gzNF2d53h2oCKDnojLtSP.eval",
      "sourceId": "2gzNF2d53h2oCKDnojLtSP",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-30b-a3b-instruct-2507",
      "score": 0.5561868686868687,
      "bestScore": 0.5561868686868687,
      "standardError": 0.028844434239616544,
      "startedAt": "2026-08-30T22:44:04.000Z",
      "releaseDate": "2025-07-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FX3HfncZtnLKUPoUuRc9YsR.eval",
      "sourceId": "X3HfncZtnLKUPoUuRc9YsR",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Qwen3-30B-A3B_none",
      "score": 0.5037878787878788,
      "bestScore": 0.5037878787878788,
      "standardError": 0.02827514183647295,
      "startedAt": "2026-08-30T22:43:32.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDGSmhVxWsi7chsuKS64VLg.eval",
      "sourceId": "DGSmhVxWsi7chsuKS64VLg",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Qwen3-30B-A3B",
      "score": 0.6167929292929293,
      "bestScore": 0.6167929292929293,
      "standardError": 0.029727711361608088,
      "startedAt": "2026-08-30T22:43:15.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FBnX67AoZzgVbHgaM8F8Brz.eval",
      "sourceId": "BnX67AoZzgVbHgaM8F8Brz",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen2.5-7b-instruct",
      "score": 0.3547979797979798,
      "bestScore": 0.3547979797979798,
      "standardError": 0.02291485990511179,
      "startedAt": "2026-08-30T22:42:14.000Z",
      "releaseDate": "2024-09-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6jkd5nJzspGZJtvZynZZTD.eval",
      "sourceId": "6jkd5nJzspGZJtvZynZZTD",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "mistral-small-3.1-2503",
      "score": 0.41919191919191917,
      "bestScore": 0.41919191919191917,
      "standardError": 0.02348827340432996,
      "startedAt": "2026-08-30T22:40:44.000Z",
      "releaseDate": "2025-03-17",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDvbGrBaybHM3UhJwWhE3LR.eval",
      "sourceId": "DvbGrBaybHM3UhJwWhE3LR",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "mistral-small-3-2501",
      "score": 0.47285353535353536,
      "bestScore": 0.47285353535353536,
      "standardError": 0.027888880246470463,
      "startedAt": "2026-08-30T22:40:23.000Z",
      "releaseDate": "2025-01-30",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYJPwhEJa6qQWB23HQjByxt.eval",
      "sourceId": "YJPwhEJa6qQWB23HQjByxt",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "magistral-small-2509",
      "score": 0.476010101010101,
      "bestScore": 0.476010101010101,
      "standardError": 0.025993462698453398,
      "startedAt": "2026-08-30T22:39:36.000Z",
      "releaseDate": "2025-09-18",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F3UbYVQELDHAcQgu7sTTnRf.eval",
      "sourceId": "3UbYVQELDHAcQgu7sTTnRf",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Llama-3.2-1B-Instruct",
      "score": 0.23926767676767677,
      "bestScore": 0.23926767676767677,
      "standardError": 0.016278803912773237,
      "startedAt": "2026-08-30T22:39:11.000Z",
      "releaseDate": "2024-09-24",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFGEsLTxzZ7fE9aq7DhdqZb.eval",
      "sourceId": "FGEsLTxzZ7fE9aq7DhdqZb",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "glm-4.7-flash_none",
      "score": 0.4513888888888889,
      "bestScore": 0.4513888888888889,
      "standardError": 0.021751012010888356,
      "startedAt": "2026-08-30T22:37:57.000Z",
      "releaseDate": null,
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGSde7u2Mb6GZGpubJjjPam.eval",
      "sourceId": "GSde7u2Mb6GZGpubJjjPam",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "glm-4.7-flash",
      "score": 0.6054292929292929,
      "bestScore": 0.6054292929292929,
      "standardError": 0.02934798760577577,
      "startedAt": "2026-08-30T22:37:35.000Z",
      "releaseDate": null,
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGL7htq7XRnbzqpAVK5f862.eval",
      "sourceId": "GL7htq7XRnbzqpAVK5f862",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-6-astra_max",
      "score": 0.9577020202020202,
      "bestScore": 0.9577020202020202,
      "standardError": 0.013696812097020311,
      "startedAt": "2026-08-30T14:57:43.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "3h8STJpG2VzY9LiBUzwotv",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "granite-4.0-1b",
      "score": 0.2398989898989899,
      "bestScore": 0.2398989898989899,
      "standardError": 0.01619451031298072,
      "startedAt": "2026-08-28T19:33:14.000Z",
      "releaseDate": "2025-10-28",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FoWiE4CniV7k5adL5wDh5r8.eval",
      "sourceId": "oWiE4CniV7k5adL5wDh5r8",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.5-35b-a3b_none",
      "score": 0.8118686868686869,
      "bestScore": 0.8118686868686869,
      "standardError": 0.023198629229768078,
      "startedAt": "2026-08-28T18:43:36.000Z",
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKKm5DbPUgXo88eYr64g4SK.eval",
      "sourceId": "KKm5DbPUgXo88eYr64g4SK",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.5-35b-a3b",
      "score": 0.8345959595959596,
      "bestScore": 0.8345959595959596,
      "standardError": 0.02179018174848238,
      "startedAt": "2026-08-28T18:38:38.000Z",
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRWjuCuVGvaRd4vDjJJ6aP2.eval",
      "sourceId": "RWjuCuVGvaRd4vDjJJ6aP2",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-4b-instruct-2507",
      "score": 0.4583333333333333,
      "bestScore": 0.4583333333333333,
      "standardError": 0.028502193235395257,
      "startedAt": "2026-08-28T17:47:35.000Z",
      "releaseDate": "2025-08-06",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVhQBa9AViRt2UKSNpEAyjN.eval",
      "sourceId": "VhQBa9AViRt2UKSNpEAyjN",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-4b_none",
      "score": 0.4356060606060606,
      "bestScore": 0.4356060606060606,
      "standardError": 0.026892896738900395,
      "startedAt": "2026-08-28T17:41:08.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSLJyD8yew4TxcmprJfVvuE.eval",
      "sourceId": "SLJyD8yew4TxcmprJfVvuE",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-4b",
      "score": 0.5233585858585859,
      "bestScore": 0.5233585858585859,
      "standardError": 0.030102242301033352,
      "startedAt": "2026-08-28T17:40:19.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FG3dinPvwrci2Nu3m9pby2i.eval",
      "sourceId": "G3dinPvwrci2Nu3m9pby2i",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Qwen3-32B",
      "score": 0.6571969696969697,
      "bestScore": 0.6571969696969697,
      "standardError": 0.027890331058144285,
      "startedAt": "2026-08-28T17:33:22.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFfoxqm7YsZr3FimCGsGAJv.eval",
      "sourceId": "Ffoxqm7YsZr3FimCGsGAJv",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-14b_none",
      "score": 0.5340909090909091,
      "bestScore": 0.5340909090909091,
      "standardError": 0.02752462335115392,
      "startedAt": "2026-08-28T16:54:57.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGRQV9Q8gHoHnSEPWofiQuq.eval",
      "sourceId": "GRQV9Q8gHoHnSEPWofiQuq",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-14b",
      "score": 0.6376262626262627,
      "bestScore": 0.6376262626262627,
      "standardError": 0.02931953778086544,
      "startedAt": "2026-08-28T16:53:49.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXGK9DjJHHVWr2LyT298ooL.eval",
      "sourceId": "XGK9DjJHHVWr2LyT298ooL",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-1.7b_none",
      "score": 0.3061868686868687,
      "bestScore": 0.3061868686868687,
      "standardError": 0.022564006554294688,
      "startedAt": "2026-08-28T16:47:40.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGdfzaHmiZht2DKPZDtQAET.eval",
      "sourceId": "GdfzaHmiZht2DKPZDtQAET",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-1.7b",
      "score": 0.380050505050505,
      "bestScore": 0.380050505050505,
      "standardError": 0.02748387280745701,
      "startedAt": "2026-08-28T16:47:04.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FF3CtjUceBrMi9iYmjULx7J.eval",
      "sourceId": "F3CtjUceBrMi9iYmjULx7J",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "mistral-small-3.2-2506",
      "score": 0.49053030303030304,
      "bestScore": 0.49053030303030304,
      "standardError": 0.027717443261958633,
      "startedAt": "2026-08-28T16:15:28.000Z",
      "releaseDate": "2025-06-20",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FV2CVKXcEdLnEXmppjUEB9M.eval",
      "sourceId": "V2CVKXcEdLnEXmppjUEB9M",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "granite-4.0-350m",
      "score": 0.11237373737373738,
      "bestScore": 0.11237373737373738,
      "standardError": 0.009083571075486867,
      "startedAt": "2026-08-28T12:37:41.000Z",
      "releaseDate": "2025-10-28",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FF6xZSehzKjpybEh8JkGwni.eval",
      "sourceId": "F6xZSehzKjpybEh8JkGwni",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemma-3-4b-it",
      "score": 0.23232323232323232,
      "bestScore": 0.23232323232323232,
      "standardError": 0.018690142324718148,
      "startedAt": "2026-08-28T08:35:26.000Z",
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMcJyXLHBzx5VjAqXjmrrsb.eval",
      "sourceId": "McJyXLHBzx5VjAqXjmrrsb",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemma-3-1b-it",
      "score": 0.1994949494949495,
      "bestScore": 0.1994949494949495,
      "standardError": 0.01521201537793133,
      "startedAt": "2026-08-28T08:27:48.000Z",
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPc6CyTEYKbcdDUDKzGZnS9.eval",
      "sourceId": "Pc6CyTEYKbcdDUDKzGZnS9",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemma-3-12b-it",
      "score": 0.39457070707070707,
      "bestScore": 0.39457070707070707,
      "standardError": 0.025254327795483048,
      "startedAt": "2026-08-28T08:22:17.000Z",
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKfQKr2ibCFMx5xBELoiM3Y.eval",
      "sourceId": "KfQKr2ibCFMx5xBELoiM3Y",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "DeepSeek-R1-Distill-Qwen-32B",
      "score": 0.6414141414141414,
      "bestScore": 0.6414141414141414,
      "standardError": 0.027534177016461762,
      "startedAt": "2026-08-28T08:15:19.000Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fdwyb2yzFewGT6w6oFdVMAc.eval",
      "sourceId": "dwyb2yzFewGT6w6oFdVMAc",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "DeepSeek-R1-Distill-Qwen-1.5B",
      "score": 0.33585858585858586,
      "bestScore": 0.33585858585858586,
      "standardError": 0.020592418168039042,
      "startedAt": "2026-08-28T07:52:16.000Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FESnXA8LqMqhjcsoWQaMLVQ.eval",
      "sourceId": "ESnXA8LqMqhjcsoWQaMLVQ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "deepseek-r1-0528-qwen3-8b",
      "score": 0.0928030303030303,
      "bestScore": 0.0928030303030303,
      "standardError": 0.010663485573693416,
      "startedAt": "2026-08-28T07:48:47.000Z",
      "releaseDate": "2025-05-28",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdszRh7ELyCEukWnwYnUxMp.eval",
      "sourceId": "dszRh7ELyCEukWnwYnUxMp",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "QwQ-32B",
      "score": 0.6534090909090909,
      "bestScore": 0.6534090909090909,
      "standardError": 0.0281399081227253,
      "startedAt": "2026-08-28T07:20:47.000Z",
      "releaseDate": "2025-03-05",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPUXMxQiRY6XaG6NjwsEDLy.eval",
      "sourceId": "PUXMxQiRY6XaG6NjwsEDLy",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Llama-3.1-8B-Instruct",
      "score": 0.26957070707070707,
      "bestScore": 0.26957070707070707,
      "standardError": 0.018583813167150035,
      "startedAt": "2026-08-27T16:30:31.000Z",
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfxwRTtt8bLV4DNutwtMEox.eval",
      "sourceId": "fxwRTtt8bLV4DNutwtMEox",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.5-9B_none",
      "score": 0.7891414141414141,
      "bestScore": 0.7891414141414141,
      "standardError": 0.023888485629911964,
      "startedAt": "2026-08-27T14:30:46.000Z",
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FAczFCvkWR5SmdbV7eXruqj.eval",
      "sourceId": "AczFCvkWR5SmdbV7eXruqj",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.5-9B",
      "score": 0.7897727272727273,
      "bestScore": 0.7897727272727273,
      "standardError": 0.02365568030958479,
      "startedAt": "2026-08-27T14:30:17.000Z",
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fo6migB5WP27uLr9nUWpCiM.eval",
      "sourceId": "o6migB5WP27uLr9nUWpCiM",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-8b_none",
      "score": 0.47474747474747475,
      "bestScore": 0.47474747474747475,
      "standardError": 0.02623286175358367,
      "startedAt": "2026-08-27T14:21:49.000Z",
      "releaseDate": "2025-04-28",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLcttjHHzTyAW6MqNE29xiK.eval",
      "sourceId": "LcttjHHzTyAW6MqNE29xiK",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-8b",
      "score": 0.5675505050505051,
      "bestScore": 0.5675505050505051,
      "standardError": 0.029327713464985333,
      "startedAt": "2026-08-27T14:21:43.000Z",
      "releaseDate": "2025-04-28",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjRiWmBxtBAyiXiRBRMY8wX.eval",
      "sourceId": "jRiWmBxtBAyiXiRBRMY8wX",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemma-3-27b-it",
      "score": 0.4772727272727273,
      "bestScore": 0.4772727272727273,
      "standardError": 0.026982120891168943,
      "startedAt": "2026-08-27T14:17:43.000Z",
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjdFH2zZdssYKDVJQPKXvNB.eval",
      "sourceId": "jdFH2zZdssYKDVJQPKXvNB",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-oss-20b_medium",
      "score": 0.6079545454545454,
      "bestScore": 0.6079545454545454,
      "standardError": 0.024864059770258008,
      "startedAt": "2026-08-27T14:06:50.000Z",
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGEwyLqZT93Hp55fdtHrJhm.eval",
      "sourceId": "GEwyLqZT93Hp55fdtHrJhm",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-oss-20b_low",
      "score": 0.5315656565656566,
      "bestScore": 0.5315656565656566,
      "standardError": 0.02502509361882172,
      "startedAt": "2026-08-27T14:05:49.000Z",
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDCJQMqipndGxwMXe8BDGeg.eval",
      "sourceId": "DCJQMqipndGxwMXe8BDGeg",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "glm-5.3-flash_max",
      "score": 0.9015151515151515,
      "bestScore": 0.9015151515151515,
      "standardError": 0.017183231147411448,
      "startedAt": "2026-08-26T22:26:55.000Z",
      "releaseDate": "2026-08-20",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "Jvt5BNr4ERbE9QGgMcEbmT",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "glm-5.3_max",
      "score": 0.9090909090909091,
      "bestScore": 0.9090909090909091,
      "standardError": 0.015851578202142853,
      "startedAt": "2026-08-24T19:44:18.000Z",
      "releaseDate": "2026-08-14",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "dLDBahYBTGRNjbsXA5pyr3",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "deepseek-v4-pro-0813_max",
      "score": 0.9166666666666666,
      "bestScore": 0.9166666666666666,
      "standardError": 0.015242573629934338,
      "startedAt": "2026-08-18T21:03:23.000Z",
      "releaseDate": "2026-08-13",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "nEJGtXGN588zynPkynZkxz",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.7-flash_high",
      "score": 0.9482323232323232,
      "bestScore": 0.9482323232323232,
      "standardError": 0.013456525292246259,
      "startedAt": "2026-08-14T18:29:12.000Z",
      "releaseDate": "2026-08-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XyVL3VffJ6UboY97vxpohL",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "inkling-small_xhigh",
      "score": 0.88510101010101,
      "bestScore": 0.88510101010101,
      "standardError": 0.019100012831823398,
      "startedAt": "2026-08-14T14:19:47.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "jTHw5SET8usd7emYPjEDjS",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "grok-4.6_xhigh",
      "score": 0.9318181818181818,
      "bestScore": 0.9318181818181818,
      "standardError": 0.015204299687218694,
      "startedAt": "2026-08-14T13:47:42.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9YJf2nviwjM6WVhw8qeJta",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "grok-4.6_high",
      "score": 0.9400252525252525,
      "bestScore": 0.9400252525252525,
      "standardError": 0.014480018812911388,
      "startedAt": "2026-08-12T19:39:52.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hnP6Z88x8bPEBRZ2PW6kps",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "nemotron-3-ultra",
      "score": 0.8535353535353535,
      "bestScore": 0.8535353535353535,
      "standardError": 0.025190921114603887,
      "startedAt": "2026-08-10T17:46:52.000Z",
      "releaseDate": "2026-06-04",
      "organization": "Nvidia",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbpZy8tAbh7UV6A6U7XZtBg.eval",
      "sourceId": "bpZy8tAbh7UV6A6U7XZtBg",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "MiniMax-M3_none",
      "score": 0.8131313131313131,
      "bestScore": 0.8131313131313131,
      "standardError": 0.027772533334218946,
      "startedAt": "2026-08-10T17:46:29.000Z",
      "releaseDate": "2026-06-01",
      "organization": "MiniMax",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKM4G49tQZB8ce67RxYGyMy.eval",
      "sourceId": "KM4G49tQZB8ce67RxYGyMy",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "MiniMax-M3",
      "score": 0.9090909090909091,
      "bestScore": 0.9090909090909091,
      "standardError": 0.020482086775424225,
      "startedAt": "2026-08-10T17:46:03.000Z",
      "releaseDate": "2026-06-01",
      "organization": "MiniMax",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhDESmyGFrbZDomjff3rqzD.eval",
      "sourceId": "hDESmyGFrbZDomjff3rqzD",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "glm-5.2_none",
      "score": 0.7121212121212122,
      "bestScore": 0.7121212121212122,
      "standardError": 0.03225883512300997,
      "startedAt": "2026-08-10T17:45:32.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYDQJAMgMjLaikPkeVDbAjr.eval",
      "sourceId": "YDQJAMgMjLaikPkeVDbAjr",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "glm-5.2_low",
      "score": 0.8787878787878788,
      "bestScore": 0.8787878787878788,
      "standardError": 0.02325315795194205,
      "startedAt": "2026-08-10T17:45:10.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSB6Qj6rCBDk8jZmFcCTzWc.eval",
      "sourceId": "SB6Qj6rCBDk8jZmFcCTzWc",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "glm-5.1",
      "score": 0.898989898989899,
      "bestScore": 0.898989898989899,
      "standardError": 0.021469735576055356,
      "startedAt": "2026-08-10T17:44:34.000Z",
      "releaseDate": "2026-04-07",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCBbeMKRRSeRsXnvSfn7Png.eval",
      "sourceId": "CBbeMKRRSeRsXnvSfn7Png",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-mini-2026-03-17_xhigh",
      "score": 0.8686868686868687,
      "bestScore": 0.8686868686868687,
      "standardError": 0.02406315641682249,
      "startedAt": "2026-08-07T01:00:11.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgKB2rCD7mHzFN2MCt2YCni.eval",
      "sourceId": "gKB2rCD7mHzFN2MCt2YCni",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.6-sol_none",
      "score": 0.8282828282828283,
      "bestScore": 0.8282828282828283,
      "standardError": 0.026869716187429872,
      "startedAt": "2026-08-07T01:00:09.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDDfYPyKQW7vgDtGgkwcKSi.eval",
      "sourceId": "DDfYPyKQW7vgDtGgkwcKSi",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.5_none",
      "score": 0.7727272727272727,
      "bestScore": 0.7727272727272727,
      "standardError": 0.029857515673386438,
      "startedAt": "2026-08-07T00:57:11.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKesKa5EXtFG7H4UMBFzgCh.eval",
      "sourceId": "KesKa5EXtFG7H4UMBFzgCh",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "kimi-k3_low",
      "score": 0.8484848484848485,
      "bestScore": 0.8484848484848485,
      "standardError": 0.025545650426603637,
      "startedAt": "2026-08-07T00:54:16.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmvNb3hNgSZ2gS6jHMtvCdk.eval",
      "sourceId": "mvNb3hNgSZ2gS6jHMtvCdk",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "kimi-k3_high",
      "score": 0.9191919191919192,
      "bestScore": 0.9191919191919192,
      "standardError": 0.019417681889724515,
      "startedAt": "2026-08-07T00:53:51.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FAoAD5uVtgVDnXmwA7STuno.eval",
      "sourceId": "AoAD5uVtgVDnXmwA7STuno",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.6-terra_none",
      "score": 0.7727272727272727,
      "bestScore": 0.7727272727272727,
      "standardError": 0.029857515673386438,
      "startedAt": "2026-08-07T00:50:06.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FF7v3JtKFb27j2YVmsndLM5.eval",
      "sourceId": "F7v3JtKFb27j2YVmsndLM5",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-mini-2026-03-17_none",
      "score": 0.6414141414141414,
      "bestScore": 0.6414141414141414,
      "standardError": 0.034169036403915276,
      "startedAt": "2026-08-07T00:46:46.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQyz2Qt5x4P2XwB3w4h4jFR.eval",
      "sourceId": "Qyz2Qt5x4P2XwB3w4h4jFR",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.6-max-preview",
      "score": 0.8737373737373737,
      "bestScore": 0.8737373737373737,
      "standardError": 0.023664359402880177,
      "startedAt": "2026-08-07T00:46:39.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJv2hJvjpvA8xJW3Bzv7bfw.eval",
      "sourceId": "Jv2hJvjpvA8xJW3Bzv7bfw",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.6-sol_low",
      "score": 0.898989898989899,
      "bestScore": 0.898989898989899,
      "standardError": 0.021469735576055356,
      "startedAt": "2026-08-07T00:46:32.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6cmgkNAQZekBr6e6rVdLe9.eval",
      "sourceId": "6cmgkNAQZekBr6e6rVdLe9",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.7-max",
      "score": 0.9090909090909091,
      "bestScore": 0.9090909090909091,
      "standardError": 0.020482086775424225,
      "startedAt": "2026-08-07T00:46:28.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FW6mTanyjvXhasFBzHukv4S.eval",
      "sourceId": "W6mTanyjvXhasFBzHukv4S",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o3-2025-04-16_medium",
      "score": 0.8080808080808081,
      "bestScore": 0.8080808080808081,
      "standardError": 0.028057791672989062,
      "startedAt": "2026-08-07T00:42:14.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTBYA7JWneLUCy9kyktFBbt.eval",
      "sourceId": "TBYA7JWneLUCy9kyktFBbt",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.1-2025-11-13_none",
      "score": 0.6666666666666666,
      "bestScore": 0.6666666666666666,
      "standardError": 0.033586181457325226,
      "startedAt": "2026-08-07T00:38:38.000Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FUT5fVL4EGvSTZBKMTPy9YN.eval",
      "sourceId": "UT5fVL4EGvSTZBKMTPy9YN",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "kimi-k2.7-code",
      "score": 0.8787878787878788,
      "bestScore": 0.8787878787878788,
      "standardError": 0.02325315795194205,
      "startedAt": "2026-08-07T00:38:29.000Z",
      "releaseDate": "2026-06-12",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXneEig6pACkAWeekvgCwNk.eval",
      "sourceId": "XneEig6pACkAWeekvgCwNk",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.6-luna_none",
      "score": 0.6363636363636364,
      "bestScore": 0.6363636363636364,
      "standardError": 0.03427308652999934,
      "startedAt": "2026-08-07T00:35:08.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbDKKx7zTdW5WwMCfwqsLi4.eval",
      "sourceId": "bDKKx7zTdW5WwMCfwqsLi4",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.6-terra_low",
      "score": 0.8737373737373737,
      "bestScore": 0.8737373737373737,
      "standardError": 0.023664359402880177,
      "startedAt": "2026-08-07T00:35:04.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJcuoAWhcnngPEZBPmiUyLF.eval",
      "sourceId": "JcuoAWhcnngPEZBPmiUyLF",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.6-plus",
      "score": 0.8838383838383839,
      "bestScore": 0.8838383838383839,
      "standardError": 0.022828881775249357,
      "startedAt": "2026-08-07T00:31:39.000Z",
      "releaseDate": "2026-03-31",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Ff6ZdMtRpdfskGbrBCzrKSn.eval",
      "sourceId": "f6ZdMtRpdfskGbrBCzrKSn",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o4-mini-2025-04-16_medium",
      "score": 0.7777777777777778,
      "bestScore": 0.7777777777777778,
      "standardError": 0.02962022787479049,
      "startedAt": "2026-08-07T00:24:28.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FnoAGevebNp5W5yAnmbEYaz.eval",
      "sourceId": "noAGevebNp5W5yAnmbEYaz",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.6-27b",
      "score": 0.8585858585858586,
      "bestScore": 0.8585858585858586,
      "standardError": 0.024825909793343353,
      "startedAt": "2026-08-07T00:24:27.000Z",
      "releaseDate": "2026-04-22",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbuLf8Ne29xmzMZBRQkvrQG.eval",
      "sourceId": "buLf8Ne29xmzMZBRQkvrQG",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.6-27b_none",
      "score": 0.8484848484848485,
      "bestScore": 0.8484848484848485,
      "standardError": 0.02554565042660364,
      "startedAt": "2026-08-07T00:24:17.000Z",
      "releaseDate": "2026-04-22",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWeP7Q27N5PzxSgg4ybgijX.eval",
      "sourceId": "WeP7Q27N5PzxSgg4ybgijX",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.5-397b-a17b",
      "score": 0.8585858585858586,
      "bestScore": 0.8585858585858586,
      "standardError": 0.024825909793343353,
      "startedAt": "2026-08-07T00:24:06.000Z",
      "releaseDate": "2026-02-13",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgZMFRbVKnBmcnZU4YqqKMN.eval",
      "sourceId": "gZMFRbVKnBmcnZU4YqqKMN",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.5-397b-a17b_none",
      "score": 0.8636363636363636,
      "bestScore": 0.8636363636363636,
      "standardError": 0.02445015597318985,
      "startedAt": "2026-08-07T00:24:02.000Z",
      "releaseDate": "2026-02-13",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbXrmQqzF9sXbuZfRrSc7NP.eval",
      "sourceId": "bXrmQqzF9sXbuZfRrSc7NP",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.5-plus",
      "score": 0.8484848484848485,
      "bestScore": 0.8484848484848485,
      "standardError": 0.025545650426603637,
      "startedAt": "2026-08-07T00:20:32.000Z",
      "releaseDate": "2026-02-16",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmGXYSUNCMisvaRzhQp6g2L.eval",
      "sourceId": "mGXYSUNCMisvaRzhQp6g2L",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.6-flash_minimal",
      "score": 0.8585858585858586,
      "bestScore": 0.8585858585858586,
      "standardError": 0.024825909793343353,
      "startedAt": "2026-08-07T00:17:01.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FaPmR8fV49N98a6NVKpASAf.eval",
      "sourceId": "aPmR8fV49N98a6NVKpASAf",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.6-35b-a3b",
      "score": 0.8383838383838383,
      "bestScore": 0.8383838383838383,
      "standardError": 0.02622591986362927,
      "startedAt": "2026-08-07T00:12:45.000Z",
      "releaseDate": "2026-04-14",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FEWAyxFsRUvuu4NpmmLqWkn.eval",
      "sourceId": "EWAyxFsRUvuu4NpmmLqWkn",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.6-35b-a3b_none",
      "score": 0.8484848484848485,
      "bestScore": 0.8484848484848485,
      "standardError": 0.02554565042660364,
      "startedAt": "2026-08-07T00:12:39.000Z",
      "releaseDate": "2026-04-14",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9kErpmgQ42wTvgJaFn478v.eval",
      "sourceId": "9kErpmgQ42wTvgJaFn478v",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.6-flash",
      "score": 0.8333333333333334,
      "bestScore": 0.8333333333333334,
      "standardError": 0.026552207828215258,
      "startedAt": "2026-08-07T00:11:57.000Z",
      "releaseDate": "2026-04-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FH2FvRXagQRQmiNa7WrNMU2.eval",
      "sourceId": "H2FvRXagQRQmiNa7WrNMU2",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.6-luna_low",
      "score": 0.8232323232323232,
      "bestScore": 0.8232323232323232,
      "standardError": 0.027178752639044908,
      "startedAt": "2026-08-07T00:08:31.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FnW39ZzuDQ3xTAGVoEfFG95.eval",
      "sourceId": "nW39ZzuDQ3xTAGVoEfFG95",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.7-plus",
      "score": 0.8787878787878788,
      "bestScore": 0.8787878787878788,
      "standardError": 0.02325315795194205,
      "startedAt": "2026-08-07T00:07:25.000Z",
      "releaseDate": "2026-06-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FURoWjw9Htc6QPwABVcrybw.eval",
      "sourceId": "URoWjw9Htc6QPwABVcrybw",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.7-plus_none",
      "score": 0.8181818181818182,
      "bestScore": 0.8181818181818182,
      "standardError": 0.027479603010538825,
      "startedAt": "2026-08-07T00:03:42.000Z",
      "releaseDate": "2026-06-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FBWz8uig4QyqPtke59pM88z.eval",
      "sourceId": "BWz8uig4QyqPtke59pM88z",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.7-flash",
      "score": 0.8232323232323232,
      "bestScore": 0.8232323232323232,
      "standardError": 0.027178752639044908,
      "startedAt": "2026-08-07T00:03:33.000Z",
      "releaseDate": "2026-07-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fi6i3gQf3ng2aAipvowPpws.eval",
      "sourceId": "i6i3gQf3ng2aAipvowPpws",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.7-flash_none",
      "score": 0.8080808080808081,
      "bestScore": 0.8080808080808081,
      "standardError": 0.028057791672989062,
      "startedAt": "2026-08-07T00:03:20.000Z",
      "releaseDate": "2026-07-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYhds8GYkCXWgRQhuLiURua.eval",
      "sourceId": "Yhds8GYkCXWgRQhuLiURua",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-nano-2026-03-17_none",
      "score": 0.5555555555555556,
      "bestScore": 0.5555555555555556,
      "standardError": 0.035402943770953675,
      "startedAt": "2026-08-07T00:01:43.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgLewKMc7ttCD2moUrgAyYX.eval",
      "sourceId": "gLewKMc7ttCD2moUrgAyYX",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.5-flash",
      "score": 0.8232323232323232,
      "bestScore": 0.8232323232323232,
      "standardError": 0.027178752639044908,
      "startedAt": "2026-08-07T00:00:59.000Z",
      "releaseDate": "2026-02-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FS7wbot7JR53BAjM8F9gaws.eval",
      "sourceId": "S7wbot7JR53BAjM8F9gaws",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5-mini-2025-08-07_minimal",
      "score": 0.7171717171717171,
      "bestScore": 0.7171717171717171,
      "standardError": 0.032087795587867514,
      "startedAt": "2026-08-07T00:00:44.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhYSstHWNUNwYoZdmZpYQqK.eval",
      "sourceId": "hYSstHWNUNwYoZdmZpYQqK",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-nano-2026-03-17_low",
      "score": 0.7222222222222222,
      "bestScore": 0.7222222222222222,
      "standardError": 0.03191178226713548,
      "startedAt": "2026-08-07T00:00:23.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FNq7w5iKALDSuT5oQdn5m8f.eval",
      "sourceId": "Nq7w5iKALDSuT5oQdn5m8f",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemma-4-31b-it_minimal",
      "score": 0.7575757575757576,
      "bestScore": 0.7575757575757576,
      "standardError": 0.03053289223393202,
      "startedAt": "2026-08-06T23:59:38.000Z",
      "releaseDate": "2026-04-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FP5L8qbXo42vTvbU3UvTNjk.eval",
      "sourceId": "P5L8qbXo42vTvbU3UvTNjk",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemma-4-26b-a4b-it_minimal",
      "score": 0.7323232323232324,
      "bestScore": 0.7323232323232324,
      "standardError": 0.031544498882702825,
      "startedAt": "2026-08-06T23:59:25.000Z",
      "releaseDate": "2026-04-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FUsx9KaiXvHZU4qrzgc4sLg.eval",
      "sourceId": "Usx9KaiXvHZU4qrzgc4sLg",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "deepseek-v4-pro_none",
      "score": 0.7323232323232324,
      "bestScore": 0.7323232323232324,
      "standardError": 0.031544498882702825,
      "startedAt": "2026-08-06T23:58:34.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFtGueFbnFnKBAuLmEEufFg.eval",
      "sourceId": "FtGueFbnFnKBAuLmEEufFg",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "deepseek-v4-pro_high",
      "score": 0.9090909090909091,
      "bestScore": 0.9090909090909091,
      "standardError": 0.020482086775424225,
      "startedAt": "2026-08-06T23:58:27.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTPdRwHTJtaJYSu2Cxc5Czn.eval",
      "sourceId": "TPdRwHTJtaJYSu2Cxc5Czn",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-6_max",
      "score": 0.8838383838383839,
      "bestScore": 0.8838383838383839,
      "standardError": 0.02282888177524936,
      "startedAt": "2026-08-06T23:57:58.000Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJcQMgXdZKU54u52B8TGQKp.eval",
      "sourceId": "JcQMgXdZKU54u52B8TGQKp",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-oss-20b_high",
      "score": 0.4595959595959596,
      "bestScore": 0.4595959595959596,
      "standardError": 0.03550702465131341,
      "startedAt": "2026-08-06T22:50:56.000Z",
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhYATFPeUnMq7Std2nLUVAJ.eval",
      "sourceId": "hYATFPeUnMq7Std2nLUVAJ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.6-flash_low",
      "score": 0.8636363636363636,
      "bestScore": 0.8636363636363636,
      "standardError": 0.02445015597318985,
      "startedAt": "2026-08-06T22:36:03.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJoraLb9y55QXpMg4Rrmr5x.eval",
      "sourceId": "JoraLb9y55QXpMg4Rrmr5x",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.5-flash-lite_low",
      "score": 0.7575757575757576,
      "bestScore": 0.7575757575757576,
      "standardError": 0.030532892233932022,
      "startedAt": "2026-08-06T22:35:12.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdgSJTFQ4UtuQKEhB9ss6K7.eval",
      "sourceId": "dgSJTFQ4UtuQKEhB9ss6K7",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.5-flash-lite_minimal",
      "score": 0.7424242424242424,
      "bestScore": 0.7424242424242424,
      "standardError": 0.031156269519646826,
      "startedAt": "2026-08-06T22:35:11.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7aAKcnsCXoSumfAu8mr2un.eval",
      "sourceId": "7aAKcnsCXoSumfAu8mr2un",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.5-flash-lite_high",
      "score": 0.8333333333333334,
      "bestScore": 0.8333333333333334,
      "standardError": 0.026552207828215258,
      "startedAt": "2026-08-06T22:35:04.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fchbh6XstiqtUa5W9zuyfjS.eval",
      "sourceId": "chbh6XstiqtUa5W9zuyfjS",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.5-flash_low",
      "score": 0.8888888888888888,
      "bestScore": 0.8888888888888888,
      "standardError": 0.02239078763821682,
      "startedAt": "2026-08-06T22:30:55.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FeiTYduCDNq3HgwTg4ZhEw6.eval",
      "sourceId": "eiTYduCDNq3HgwTg4ZhEw6",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.1-pro-preview_high",
      "score": 0.9444444444444444,
      "bestScore": 0.9444444444444444,
      "standardError": 0.016319950700767385,
      "startedAt": "2026-08-06T22:30:37.000Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FBgmrBFsbsaD8rNpebv49Da.eval",
      "sourceId": "BgmrBFsbsaD8rNpebv49Da",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.1-flash-lite_minimal",
      "score": 0.7373737373737373,
      "bestScore": 0.7373737373737373,
      "standardError": 0.031353050095330834,
      "startedAt": "2026-08-06T22:26:14.000Z",
      "releaseDate": "2026-03-03",
      "organization": "Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCmHy8cVECGkvQTwakfjRo2.eval",
      "sourceId": "CmHy8cVECGkvQTwakfjRo2",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.1-flash-lite_high",
      "score": 0.8181818181818182,
      "bestScore": 0.8181818181818182,
      "standardError": 0.027479603010538825,
      "startedAt": "2026-08-06T22:26:07.000Z",
      "releaseDate": "2026-03-03",
      "organization": "Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FnE4hmJo3QV6X5jeGWAeAze.eval",
      "sourceId": "nE4hmJo3QV6X5jeGWAeAze",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.1-flash-lite_low",
      "score": 0.7424242424242424,
      "bestScore": 0.7424242424242424,
      "standardError": 0.03115626951964683,
      "startedAt": "2026-08-06T22:25:58.000Z",
      "releaseDate": "2026-03-03",
      "organization": "Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F99ys2jgQiPQ5UwrqnazfqA.eval",
      "sourceId": "99ys2jgQiPQ5UwrqnazfqA",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3-flash-preview_high",
      "score": 0.8939393939393939,
      "bestScore": 0.8939393939393939,
      "standardError": 0.021938047738853106,
      "startedAt": "2026-08-06T22:22:58.000Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgVdraDT3pXmaExP9VbqME4.eval",
      "sourceId": "gVdraDT3pXmaExP9VbqME4",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-5_max",
      "score": 0.803030303030303,
      "bestScore": 0.803030303030303,
      "standardError": 0.028335609732463334,
      "startedAt": "2026-08-06T22:21:47.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFNXAt76ZUS3sWhQDFMga7U.eval",
      "sourceId": "FNXAt76ZUS3sWhQDFMga7U",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-6_max",
      "score": 0.7878787878787878,
      "bestScore": 0.7878787878787878,
      "standardError": 0.029126522834586777,
      "startedAt": "2026-08-06T22:21:09.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSEePRKRcv8oqn5HpD2KFxB.eval",
      "sourceId": "SEePRKRcv8oqn5HpD2KFxB",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-5",
      "score": 0.9292929292929293,
      "bestScore": 0.9292929292929293,
      "standardError": 0.018263105420199485,
      "startedAt": "2026-08-06T22:21:07.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmLmUnBuX6DYKtb6x46GteM.eval",
      "sourceId": "mLmUnBuX6DYKtb6x46GteM",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-5_low",
      "score": 0.8787878787878788,
      "bestScore": 0.8787878787878788,
      "standardError": 0.02325315795194205,
      "startedAt": "2026-08-06T22:21:06.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWfaJS4L7Xdz5ZBXgJMbRGb.eval",
      "sourceId": "WfaJS4L7Xdz5ZBXgJMbRGb",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-8_none",
      "score": 0.8535353535353535,
      "bestScore": 0.8535353535353535,
      "standardError": 0.025190921114603887,
      "startedAt": "2026-08-06T22:19:27.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FckbZdkrXPwbFdjn6bmUQsi.eval",
      "sourceId": "ckbZdkrXPwbFdjn6bmUQsi",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-8_low",
      "score": 0.8838383838383839,
      "bestScore": 0.8838383838383839,
      "standardError": 0.022828881775249357,
      "startedAt": "2026-08-06T22:19:23.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGThtce6GAmRAS8srQyGQoA.eval",
      "sourceId": "GThtce6GAmRAS8srQyGQoA",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-7_max",
      "score": 0.8636363636363636,
      "bestScore": 0.8636363636363636,
      "standardError": 0.02445015597318985,
      "startedAt": "2026-08-06T22:19:01.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJu7pUkkwYVj6ugi9Ri5Jjp.eval",
      "sourceId": "Ju7pUkkwYVj6ugi9Ri5Jjp",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-fable-5_max",
      "score": 0.8585858585858586,
      "bestScore": 0.8585858585858586,
      "standardError": 0.024825909793343353,
      "startedAt": "2026-08-06T22:17:56.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FS5pm7ctk9zgPfbyHrbD9AV.eval",
      "sourceId": "S5pm7ctk9zgPfbyHrbD9AV",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-fable-5_low",
      "score": 0.7878787878787878,
      "bestScore": 0.7878787878787878,
      "standardError": 0.029126522834586777,
      "startedAt": "2026-08-06T22:17:37.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FU5PsUdX2ak4kJ2hgnQkWnX.eval",
      "sourceId": "U5PsUdX2ak4kJ2hgnQkWnX",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-fable-5_high",
      "score": 0.8333333333333334,
      "bestScore": 0.8333333333333334,
      "standardError": 0.026552207828215258,
      "startedAt": "2026-08-06T22:17:31.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fd23hYvSNGFo7rniLFyBWiu.eval",
      "sourceId": "d23hYvSNGFo7rniLFyBWiu",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "magistral-small-2506",
      "score": 0.5606060606060606,
      "bestScore": 0.5606060606060606,
      "standardError": 0.0353608594752948,
      "startedAt": "2026-08-06T22:17:01.000Z",
      "releaseDate": "2025-06-10",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdwGLob8Vhrwx4Wn376jVmt.eval",
      "sourceId": "dwGLob8Vhrwx4Wn376jVmt",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Inkling_xhigh",
      "score": 0.8825757575757576,
      "bestScore": 0.8825757575757576,
      "standardError": 0.019355903133033585,
      "startedAt": "2026-08-05T18:55:39.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dfmiZmjAcNkLuAJL4z65eb",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3.8-max_xhigh",
      "score": 0.9267676767676768,
      "bestScore": 0.9267676767676768,
      "standardError": 0.01686498021450146,
      "startedAt": "2026-08-04T16:28:46.000Z",
      "releaseDate": "2026-08-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "TEVNDC9QXBh7PQvX9kHZip",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.5-instant",
      "score": 0.8251262626262627,
      "bestScore": 0.8251262626262627,
      "standardError": 0.022532782662252126,
      "startedAt": "2026-08-02T02:46:37.000Z",
      "releaseDate": "2026-05-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "WPVhgi8NT9grAT7kfipT9G",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "deepseek-v4-flash-0731_max",
      "score": 0.9103535353535354,
      "bestScore": 0.9103535353535354,
      "standardError": 0.017655466909172534,
      "startedAt": "2026-08-02T00:22:31.000Z",
      "releaseDate": "2026-07-31",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "EDFusDaaFkh5n2MqKNxMBD",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.6-flash_high",
      "score": 0.9412878787878788,
      "bestScore": 0.9412878787878788,
      "standardError": 0.014028967942064968,
      "startedAt": "2026-08-02T00:21:27.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "86SY5JbPB49xXhHFQW6tHu",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-5_max",
      "score": 0.9387626262626263,
      "bestScore": 0.9387626262626263,
      "standardError": 0.014782015056089346,
      "startedAt": "2026-07-24T18:40:39.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "D3mtasTumfvjR4fjmnHWRy",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o1-2024-12-17_low",
      "score": 0.7424242424242424,
      "bestScore": 0.7424242424242424,
      "standardError": 0.031156269519646826,
      "startedAt": "2026-07-20T18:29:17.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhP6P5e9zuHPToaQTTQC9A2.eval",
      "sourceId": "hP6P5e9zuHPToaQTTQC9A2",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5-2025-08-07_minimal",
      "score": 0.7171717171717171,
      "bestScore": 0.7171717171717171,
      "standardError": 0.032087795587867514,
      "startedAt": "2026-07-20T18:29:01.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fn4wyt7wHjhDdjrGacE9smk.eval",
      "sourceId": "n4wyt7wHjhDdjrGacE9smk",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o3-mini-2025-01-31_low",
      "score": 0.6818181818181818,
      "bestScore": 0.6818181818181818,
      "standardError": 0.0331847733384533,
      "startedAt": "2026-07-20T18:28:59.000Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhJf7Hxg4tgdftQ2QDNHMJA.eval",
      "sourceId": "hJf7Hxg4tgdftQ2QDNHMJA",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "kimi-k3_max",
      "score": 0.9311868686868687,
      "bestScore": 0.9311868686868687,
      "standardError": 0.014937225364194858,
      "startedAt": "2026-07-16T21:59:40.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "B8MaPUe3EYwUZASsNqE7C9",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "deepseek-chat",
      "score": 0.7121212121212122,
      "bestScore": 0.7121212121212122,
      "standardError": 0.03225883512300997,
      "startedAt": "2026-07-16T00:10:43.000Z",
      "releaseDate": "2025-12-01",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6osoAGM4euiAAksriYyLtj.eval",
      "sourceId": "6osoAGM4euiAAksriYyLtj",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o3-2025-04-16_low",
      "score": 0.797979797979798,
      "bestScore": 0.797979797979798,
      "standardError": 0.028606204289229834,
      "startedAt": "2026-07-15T03:15:08.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FAZf6N4kygJGHPsSUkYtCba.eval",
      "sourceId": "AZf6N4kygJGHPsSUkYtCba",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-2026-03-05_medium",
      "score": 0.8888888888888888,
      "bestScore": 0.8888888888888888,
      "standardError": 0.02239078763821682,
      "startedAt": "2026-07-15T03:14:58.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FManJk43Kf9VuXowNqoDrov.eval",
      "sourceId": "ManJk43Kf9VuXowNqoDrov",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-2026-03-05_none",
      "score": 0.7474747474747475,
      "bestScore": 0.7474747474747475,
      "standardError": 0.030954055470365872,
      "startedAt": "2026-07-15T03:14:58.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLaBuzZExRqocD4FET6yx26.eval",
      "sourceId": "LaBuzZExRqocD4FET6yx26",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-2026-03-05_high",
      "score": 0.898989898989899,
      "bestScore": 0.898989898989899,
      "standardError": 0.021469735576055356,
      "startedAt": "2026-07-15T03:14:50.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFbhcT8NnJUntcyNGubJXY6.eval",
      "sourceId": "FbhcT8NnJUntcyNGubJXY6",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-2026-03-05_low",
      "score": 0.8484848484848485,
      "bestScore": 0.8484848484848485,
      "standardError": 0.025545650426603637,
      "startedAt": "2026-07-15T03:14:49.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FaYgDc2iSDpmHkQfu7YjCHc.eval",
      "sourceId": "aYgDc2iSDpmHkQfu7YjCHc",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.5-flash_minimal",
      "score": 0.8636363636363636,
      "bestScore": 0.8636363636363636,
      "standardError": 0.02445015597318985,
      "startedAt": "2026-07-15T03:14:20.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQDZRoAVaEDtdS6nECEeE35.eval",
      "sourceId": "QDZRoAVaEDtdS6nECEeE35",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5-nano-2025-08-07_low",
      "score": 0.5757575757575758,
      "bestScore": 0.5757575757575758,
      "standardError": 0.03521224908841589,
      "startedAt": "2026-07-13T17:28:21.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmNZ34pXP8JN779JoHnjNQc.eval",
      "sourceId": "mNZ34pXP8JN779JoHnjNQc",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-6_high",
      "score": 0.8333333333333334,
      "bestScore": 0.8333333333333334,
      "standardError": 0.026552207828215258,
      "startedAt": "2026-07-13T17:28:15.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDmdKPD9Z9RXwYyATAoG4Uw.eval",
      "sourceId": "DmdKPD9Z9RXwYyATAoG4Uw",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5-nano-2025-08-07_minimal",
      "score": 0.48484848484848486,
      "bestScore": 0.48484848484848486,
      "standardError": 0.03560716516531066,
      "startedAt": "2026-07-13T17:28:09.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fhp4x7bhBYhkQqZoZFf4dTZ.eval",
      "sourceId": "hp4x7bhBYhkQqZoZFf4dTZ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.2-2025-12-11_none",
      "score": 0.7323232323232324,
      "bestScore": 0.7323232323232324,
      "standardError": 0.031544498882702825,
      "startedAt": "2026-07-13T17:28:07.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F4sPs53wmM9bvN8JXzsiNCk.eval",
      "sourceId": "4sPs53wmM9bvN8JXzsiNCk",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-6_medium",
      "score": 0.8333333333333334,
      "bestScore": 0.8333333333333334,
      "standardError": 0.026552207828215258,
      "startedAt": "2026-07-13T17:28:00.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fe8MXnAM8HMF8eEzYtYA7xv.eval",
      "sourceId": "e8MXnAM8HMF8eEzYtYA7xv",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "grok-4.20-0309-reasoning",
      "score": 0.8933080808080808,
      "bestScore": 0.8933080808080808,
      "standardError": 0.01863338951496632,
      "startedAt": "2026-07-13T16:45:48.000Z",
      "releaseDate": "2026-02-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "STxFZiSPAPScUuENvXBWLc",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o4-mini-2025-04-16_low",
      "score": 0.7525252525252525,
      "bestScore": 0.7525252525252525,
      "standardError": 0.030746300742124484,
      "startedAt": "2026-07-11T08:04:01.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXwRFKnuHv5qdUp9m5uTAEb.eval",
      "sourceId": "XwRFKnuHv5qdUp9m5uTAEb",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.6-terra_max",
      "score": 0.9330808080808081,
      "bestScore": 0.9330808080808081,
      "standardError": 0.015363672147634115,
      "startedAt": "2026-07-09T17:07:54.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4bLcFNu5zF6YuZf3qVKoHE",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.6-luna_max",
      "score": 0.9160353535353535,
      "bestScore": 0.9160353535353535,
      "standardError": 0.01728284355494035,
      "startedAt": "2026-07-09T13:36:49.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LWkDtkmd3TNmLzMhfZD5HW",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.6-sol_max",
      "score": 0.9349747474747475,
      "bestScore": 0.9349747474747475,
      "standardError": 0.01570109070867593,
      "startedAt": "2026-07-09T02:07:13.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "nuKCgZReCBsXKCkNj9bK37",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "grok-4.5_high",
      "score": 0.9343434343434344,
      "bestScore": 0.9343434343434344,
      "standardError": 0.014340395485738813,
      "startedAt": "2026-07-08T23:54:22.000Z",
      "releaseDate": "2026-07-08",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9kSgmUd9vh4CBXz3tJhctN",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-5_xhigh",
      "score": 0.9053030303030303,
      "bestScore": 0.9053030303030303,
      "standardError": 0.01783832131341021,
      "startedAt": "2026-07-01T01:22:14.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "HhiYxEx3Bf8YDPUh34K4DC",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "glm-5.2_max",
      "score": 0.9185606060606061,
      "bestScore": 0.9185606060606061,
      "standardError": 0.016123207533687733,
      "startedAt": "2026-06-24T20:55:39.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "c5kpB4axzmygRywWqBKFmv",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "grok-4.3_high",
      "score": 0.8882575757575758,
      "bestScore": 0.8882575757575758,
      "standardError": 0.01963840420048217,
      "startedAt": "2026-06-17T16:44:33.000Z",
      "releaseDate": "2026-04-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Vw7spGUnJyf3vuzSjTcj2F",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "deepseek-v4-pro_max",
      "score": 0.8964646464646465,
      "bestScore": 0.8964646464646465,
      "standardError": 0.017473014662966007,
      "startedAt": "2026-06-16T20:16:55.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "LuZUnhD5rBLjRw36DaqfvZ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-8_max",
      "score": 0.9103535353535354,
      "bestScore": 0.9103535353535354,
      "standardError": 0.019157125794612026,
      "startedAt": "2026-06-07T23:03:11.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dqgqP2uEPswfECCVj2Kbr9",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.5-flash_high",
      "score": 0.928030303030303,
      "bestScore": 0.928030303030303,
      "standardError": 0.016387002835275905,
      "startedAt": "2026-05-22T16:02:44.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdYYtcwENFRtm9jZTDA2p9C.eval",
      "sourceId": "dYYtcwENFRtm9jZTDA2p9C",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.5_low",
      "score": 0.9065656565656566,
      "bestScore": 0.9065656565656566,
      "standardError": 0.017532658377721708,
      "startedAt": "2026-05-05T18:00:18.624Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2CdomqtBVz2it7a2Zx2G2t.eval",
      "sourceId": "2CdomqtBVz2it7a2Zx2G2t",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "kimi-k2.6",
      "score": 0.9078282828282829,
      "bestScore": 0.9078282828282829,
      "standardError": 0.0171745162866666,
      "startedAt": "2026-05-01T17:25:36.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLjgJoDosTLkiGVT59smA7q.eval",
      "sourceId": "LjgJoDosTLkiGVT59smA7q",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.5-pro-pre-release_xhigh",
      "score": 0.9392135642135643,
      "bestScore": 0.9392135642135643,
      "standardError": 0.015832823810775415,
      "startedAt": "2026-04-24T16:01:38.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcBmhJzrC3SkfZ7uVgGVYZX.eval",
      "sourceId": "cBmhJzrC3SkfZ7uVgGVYZX",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.5-pre-release_xhigh",
      "score": 0.9400252525252525,
      "bestScore": 0.9400252525252525,
      "standardError": 0.015469771177277509,
      "startedAt": "2026-04-24T13:35:12.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFUv4EUwR78nAsdGbiwiLEr.eval",
      "sourceId": "FUv4EUwR78nAsdGbiwiLEr",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-7_xhigh",
      "score": 0.9015151515151515,
      "bestScore": 0.9015151515151515,
      "standardError": 0.018464275887293338,
      "startedAt": "2026-04-17T16:16:45.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVJezHrJ5YPCbpui8pZgcJg.eval",
      "sourceId": "VJezHrJ5YPCbpui8pZgcJg",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-mini-2026-03-17_high",
      "score": 0.8358585858585859,
      "bestScore": 0.8358585858585859,
      "standardError": 0.022274469189069883,
      "startedAt": "2026-04-15T14:03:19.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FAQF46S9hBhosKPEMy3fyB9.eval",
      "sourceId": "AQF46S9hBhosKPEMy3fyB9",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-nano-2026-03-17_high",
      "score": 0.7847,
      "bestScore": 0.7847,
      "standardError": 0.0244,
      "startedAt": "2026-04-14T15:07:20.513Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfZB3qGSohAVYzN97mniuUp.eval",
      "sourceId": "fZB3qGSohAVYzN97mniuUp",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "muse-spark",
      "score": 0.898,
      "bestScore": 0.898,
      "standardError": 0.022,
      "startedAt": "2026-04-08T16:29:51.099Z",
      "releaseDate": "2026-04-08",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "muse-spark-pre-release-gpqa",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-pro-2026-03-05_xhigh",
      "score": 0.946,
      "bestScore": 0.946,
      "standardError": 0.016,
      "startedAt": "2026-03-20T00:19:39.820Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gpt-5-4-pro-gpqa-cloud-run",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.4-2026-03-05_xhigh",
      "score": 0.933,
      "bestScore": 0.933,
      "standardError": 0.018,
      "startedAt": "2026-03-06T02:57:02.864Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfFatyce8UvpN7ZivdmrhAy.eval",
      "sourceId": "fFatyce8UvpN7ZivdmrhAy",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-6_32K",
      "score": 0.8737373737373737,
      "bestScore": 0.8737373737373737,
      "standardError": 0.020034282226331888,
      "startedAt": "2026-02-20T17:51:37.062Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F58eQmyCfPa3FufhXJFvAL2.eval",
      "sourceId": "58eQmyCfPa3FufhXJFvAL2",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3.1-pro-preview",
      "score": 0.941,
      "bestScore": 0.941,
      "standardError": 0.017,
      "startedAt": "2026-02-20T05:01:02.541Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FkR9F3oHBCnwQW4AiTPKwpd.eval",
      "sourceId": "kR9F3oHBCnwQW4AiTPKwpd",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "glm-5",
      "score": 0.8781725888324873,
      "bestScore": 0.8781725888324873,
      "standardError": 0.02336331210092387,
      "startedAt": "2026-02-12T17:19:26.257Z",
      "releaseDate": "2026-02-11",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fn4hyz75SVpwPEisKVVjvNg.eval",
      "sourceId": "n4hyz75SVpwPEisKVVjvNg",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-6_32K",
      "score": 0.9053030303030303,
      "bestScore": 0.9053030303030303,
      "standardError": 0.016916245311450143,
      "startedAt": "2026-02-06T02:17:07.741Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDiXCMhdfhcjaCUSzYWgACy.eval",
      "sourceId": "DiXCMhdfhcjaCUSzYWgACy",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-6_64K",
      "score": 0.8876262626262627,
      "bestScore": 0.8876262626262627,
      "standardError": 0.019363218328924313,
      "startedAt": "2026-02-06T02:17:05.633Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FeB9e2mtd5nRwqdFyLEsz6k.eval",
      "sourceId": "eB9e2mtd5nRwqdFyLEsz6k",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "fireworks/kimi-k2p5",
      "score": 0.876,
      "bestScore": 0.876,
      "standardError": 0.019,
      "startedAt": "2026-02-02T11:01:01.260Z",
      "releaseDate": "2026-01-27",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFmqGUq87N4JcuSbo6Uu3U5.eval",
      "sourceId": "FmqGUq87N4JcuSbo6Uu3U5",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "glm-4.7",
      "score": 0.8333333333333334,
      "bestScore": 0.8333333333333334,
      "standardError": 0.02374901666263655,
      "startedAt": "2026-01-29T20:49:26.702Z",
      "releaseDate": "2025-12-22",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6phXzd4aCbnBUSQmBsaywa.eval",
      "sourceId": "6phXzd4aCbnBUSQmBsaywa",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3-flash-preview",
      "score": 0.8320707070707071,
      "bestScore": 0.8320707070707071,
      "standardError": 0.019036353018587762,
      "startedAt": "2025-12-17T22:39:00.278Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FoAu9CZSNPZgwiwKnoNpM8Q.eval",
      "sourceId": "oAu9CZSNPZgwiwKnoNpM8Q",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "deepseek-reasoner",
      "score": 0.8342352092352093,
      "bestScore": 0.8342352092352093,
      "standardError": 0.0201537212608744,
      "startedAt": "2025-12-16T14:32:07.252Z",
      "releaseDate": "2025-12-01",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCDdjbqiYYPtEBGoZLMPM3n.eval",
      "sourceId": "CDdjbqiYYPtEBGoZLMPM3n",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.2-2025-12-11_xhigh",
      "score": 0.914,
      "bestScore": 0.914,
      "standardError": 0.018,
      "startedAt": "2025-12-13T09:04:00.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FEjo6Fy7XYxUDcCwbejfcYX.eval",
      "sourceId": "Ejo6Fy7XYxUDcCwbejfcYX",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Qwen3-235B-A22B-Thinking-2507",
      "score": 0.8005050505050505,
      "bestScore": 0.8005050505050505,
      "standardError": 0.026000933515631548,
      "startedAt": "2025-12-11T23:00:25.608Z",
      "releaseDate": "2025-07-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fgu39ndv9QxWP3NRxSUJRaX.eval",
      "sourceId": "gu39ndv9QxWP3NRxSUJRaX",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "openai/gpt-oss-120b_high",
      "score": 0.7575757575757576,
      "bestScore": 0.7575757575757576,
      "standardError": 0.027437163443883767,
      "startedAt": "2025-12-11T22:37:44.523Z",
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLC5Rhog6wt2PT84CtJ483h.eval",
      "sourceId": "LC5Rhog6wt2PT84CtJ483h",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.2-2025-12-11_high",
      "score": 0.8819444444444444,
      "bestScore": 0.8819444444444444,
      "standardError": 0.01913858277633924,
      "startedAt": "2025-12-11T20:49:22.808Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F3Q6osT8EYezayG7yfxkSpr.eval",
      "sourceId": "3Q6osT8EYezayG7yfxkSpr",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.2-2025-12-11_medium",
      "score": 0.8787878787878788,
      "bestScore": 0.8787878787878788,
      "standardError": 0.019090689343231743,
      "startedAt": "2025-12-11T20:49:17.803Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMuhtxxvALeZZ37cSG5CVcc.eval",
      "sourceId": "MuhtxxvALeZZ37cSG5CVcc",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.2-2025-12-11_low",
      "score": 0.827020202020202,
      "bestScore": 0.827020202020202,
      "standardError": 0.02265112408975899,
      "startedAt": "2025-12-11T20:25:14.724Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FoD4Jpoi75KWChqTPDaN9aB.eval",
      "sourceId": "oD4Jpoi75KWChqTPDaN9aB",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-5-20251101_16K",
      "score": 0.8547979797979798,
      "bestScore": 0.8547979797979798,
      "standardError": 0.021508147720723254,
      "startedAt": "2025-11-25T01:10:45.648Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcXkPywkhy3NnW992ri73yG.eval",
      "sourceId": "cXkPywkhy3NnW992ri73yG",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-5-20251101_32K",
      "score": 0.860479797979798,
      "bestScore": 0.860479797979798,
      "standardError": 0.020686959671393914,
      "startedAt": "2025-11-24T20:14:34.565Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKwwo3UFcPDk6iLAeSeeRaa.eval",
      "sourceId": "Kwwo3UFcPDk6iLAeSeeRaa",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-5-20251101",
      "score": 0.8068181818181818,
      "bestScore": 0.8068181818181818,
      "standardError": 0.023909479653766753,
      "startedAt": "2025-11-24T19:57:20.151Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmEkZaYpiurQ6TEMAALK3su.eval",
      "sourceId": "mEkZaYpiurQ6TEMAALK3su",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-3-pro-preview",
      "score": 0.9261363636363636,
      "bestScore": 0.9261363636363636,
      "standardError": 0.016526963574741,
      "startedAt": "2025-11-19T15:16:10.279Z",
      "releaseDate": "2025-11-18",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCg54udQGcTCnfDGX2qePLZ.eval",
      "sourceId": "Cg54udQGcTCnfDGX2qePLZ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.1-2025-11-13_medium",
      "score": 0.8503787878787878,
      "bestScore": 0.8503787878787878,
      "standardError": 0.021250422082249816,
      "startedAt": "2025-11-17T21:28:29.934Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FV3YQtSEnQ3TPZEPPgkCKQM.eval",
      "sourceId": "V3YQtSEnQ3TPZEPPgkCKQM",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-2.5-pro",
      "score": 0.8529040404040404,
      "bestScore": 0.8529040404040404,
      "standardError": 0.02117010842348584,
      "startedAt": "2025-11-16T19:00:08.124Z",
      "releaseDate": "2025-06-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcqUZ6gcq2HBnShFRWvKjGr.eval",
      "sourceId": "cqUZ6gcq2HBnShFRWvKjGr",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5.1-2025-11-13_high",
      "score": 0.8762626262626263,
      "bestScore": 0.8762626262626263,
      "standardError": 0.01875368281077595,
      "startedAt": "2025-11-13T19:52:15.039Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCSVtMksEVjBNsGJGjW4GcQ.eval",
      "sourceId": "CSVtMksEVjBNsGJGjW4GcQ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "kimi-k2-thinking-turbo",
      "score": 0.8421717171717171,
      "bestScore": 0.8421717171717171,
      "standardError": 0.020879845731985747,
      "startedAt": "2025-11-11T09:27:01.300Z",
      "releaseDate": "2025-11-06",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmuGecFpAg9wHd6gexWJ99L.eval",
      "sourceId": "muGecFpAg9wHd6gexWJ99L",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5-nano-2025-08-07_high",
      "score": 0.6944444444444444,
      "bestScore": 0.6944444444444444,
      "standardError": 0.027594508998251045,
      "startedAt": "2025-10-30T16:20:27.447Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2pSDjMSGDh4HsAxkqUuNQP.eval",
      "sourceId": "2pSDjMSGDh4HsAxkqUuNQP",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5-mini-2025-08-07_high",
      "score": 0.75,
      "bestScore": 0.75,
      "standardError": 0.022767295427113803,
      "startedAt": "2025-10-30T09:59:56.459Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjP8EyLtPFsakgnsHe2oJM6.eval",
      "sourceId": "jP8EyLtPFsakgnsHe2oJM6",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.8617424242424242,
      "bestScore": 0.8617424242424242,
      "standardError": 0.020787836099851995,
      "startedAt": "2025-10-29T22:29:36.692Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9nPyNrZxwtoT7eS4DZbQF6.eval",
      "sourceId": "9nPyNrZxwtoT7eS4DZbQF6",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-5-20250929_16K",
      "score": 0.7878787878787878,
      "bestScore": 0.7878787878787878,
      "standardError": 0.02912652283458678,
      "startedAt": "2025-10-28T13:46:39.393Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLkVuUxvsHNzpcg3sUiH7f3.eval",
      "sourceId": "LkVuUxvsHNzpcg3sUiH7f3",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-5-20250929_59K",
      "score": 0.8232323232323232,
      "bestScore": 0.8232323232323232,
      "standardError": 0.027178752639044908,
      "startedAt": "2025-10-28T13:46:37.314Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXaGuWDxPtqVvyxqdMncKFj.eval",
      "sourceId": "XaGuWDxPtqVvyxqdMncKFj",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4-0314",
      "score": 0.3573232323232323,
      "bestScore": 0.3573232323232323,
      "standardError": 0.02363099375599328,
      "startedAt": "2025-10-23T12:16:00.615Z",
      "releaseDate": "2023-03-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFkbLkW5ed8rnZGFJMevDFn.eval",
      "sourceId": "FkbLkW5ed8rnZGFJMevDFn",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-haiku-4-5-20251001_32K",
      "score": 0.7121212121212122,
      "bestScore": 0.7121212121212122,
      "standardError": 0.03225883512300997,
      "startedAt": "2025-10-22T09:43:52.881Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGEaAdb9UmoUFbEtBx7WzAR.eval",
      "sourceId": "GEaAdb9UmoUFbEtBx7WzAR",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-5-20250929_32K",
      "score": 0.817258883248731,
      "bestScore": 0.817258883248731,
      "standardError": 0.027603867020557334,
      "startedAt": "2025-10-21T15:15:58.761Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fii6JE9j57eSbuswEsJArTa.eval",
      "sourceId": "ii6JE9j57eSbuswEsJArTa",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-haiku-4-5-20251001",
      "score": 0.6047979797979798,
      "bestScore": 0.6047979797979798,
      "standardError": 0.028154535023020826,
      "startedAt": "2025-10-16T11:58:17.749Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcgnopvpB6mzsfb9yGnYK2z.eval",
      "sourceId": "cgnopvpB6mzsfb9yGnYK2z",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-max-2025-09-23",
      "score": 0.726010101010101,
      "bestScore": 0.726010101010101,
      "standardError": 0.02734528103171235,
      "startedAt": "2025-10-06T16:27:30.400Z",
      "releaseDate": "2025-09-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F46RCssznDADYeg8MuSgvsj.eval",
      "sourceId": "46RCssznDADYeg8MuSgvsj",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-5-20250929",
      "score": 0.7373737373737373,
      "bestScore": 0.7373737373737373,
      "standardError": 0.031353050095330834,
      "startedAt": "2025-09-29T18:04:57.033Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfsZmzw8jjyUgMQGwxodBwN.eval",
      "sourceId": "fsZmzw8jjyUgMQGwxodBwN",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5-mini-2025-08-07_medium",
      "score": 0.7165404040404041,
      "bestScore": 0.7165404040404041,
      "standardError": 0.023192741878083397,
      "startedAt": "2025-08-07T19:34:31.180Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCFNzodoTSVVVgp9PUXftUP.eval",
      "sourceId": "CFNzodoTSVVVgp9PUXftUP",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5-2025-08-07_medium",
      "score": 0.8535353535353535,
      "bestScore": 0.8535353535353535,
      "standardError": 0.021314825297984473,
      "startedAt": "2025-08-07T19:34:30.433Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYQCwJRoGbGAqNCQoaL6cjk.eval",
      "sourceId": "YQCwJRoGbGAqNCQoaL6cjk",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-5-nano-2025-08-07_medium",
      "score": 0.6742424242424242,
      "bestScore": 0.6742424242424242,
      "standardError": 0.026607011205052013,
      "startedAt": "2025-08-07T19:34:29.562Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FELAxkvAigvEVdMmhxiaMiy.eval",
      "sourceId": "ELAxkvAigvEVdMmhxiaMiy",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-1-20250805_27K",
      "score": 0.7676767676767676,
      "bestScore": 0.7676767676767676,
      "standardError": 0.030088629490217445,
      "startedAt": "2025-08-05T16:53:28.370Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F3obcsGCfuwYeEm9wRyqzPg.eval",
      "sourceId": "3obcsGCfuwYeEm9wRyqzPg",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-1-20250805_16K",
      "score": 0.7727272727272727,
      "bestScore": 0.7727272727272727,
      "standardError": 0.029857515673386438,
      "startedAt": "2025-08-05T16:51:47.698Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FerLgyKQs2mg2memE9SJvK5.eval",
      "sourceId": "erLgyKQs2mg2memE9SJvK5",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-1-20250805",
      "score": 0.7323232323232324,
      "bestScore": 0.7323232323232324,
      "standardError": 0.031544498882702825,
      "startedAt": "2025-08-05T16:45:23.948Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fj62bkFDjBSeFyVpnGzuzg4.eval",
      "sourceId": "j62bkFDjBSeFyVpnGzuzg4",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-2.5-pro-preview-06-05",
      "score": 0.8484848484848485,
      "bestScore": 0.8484848484848485,
      "standardError": 0.02554565042660364,
      "startedAt": "2025-06-05T16:07:50.021Z",
      "releaseDate": "2025-06-05",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQ4qTzHnjLyiCCRvNpXioUq.eval",
      "sourceId": "Q4qTzHnjLyiCCRvNpXioUq",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen3-235b-a22b",
      "score": 0.7070707070707071,
      "bestScore": 0.7070707070707071,
      "standardError": 0.027124209393506977,
      "startedAt": "2025-06-03T11:52:52.449Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FagebmxGseTvmvN4AHN8i4m.eval",
      "sourceId": "agebmxGseTvmvN4AHN8i4m",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-2.5-pro-preview-05-06",
      "score": 0.6666666666666666,
      "bestScore": 0.6666666666666666,
      "standardError": 0.033586181457325226,
      "startedAt": "2025-06-03T09:10:11.019Z",
      "releaseDate": "2025-05-06",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTT66gGiVoeYLGQMCESPrEG.eval",
      "sourceId": "TT66gGiVoeYLGQMCESPrEG",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "DeepSeek-R1-0528",
      "score": 0.7632575757575758,
      "bestScore": 0.7632575757575758,
      "standardError": 0.024023818843083005,
      "startedAt": "2025-05-29T13:30:43.463Z",
      "releaseDate": "2025-05-28",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLiRiryJsGb3h5UuxX8bmJv.eval",
      "sourceId": "LiRiryJsGb3h5UuxX8bmJv",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-3-7-sonnet-20250219_64K",
      "score": 0.7850378787878788,
      "bestScore": 0.7973484848484849,
      "standardError": 0.026506883123060064,
      "startedAt": "2025-05-26T22:53:45.927Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2t26VSNdsAE5pefQEMP25x.eval",
      "sourceId": "2t26VSNdsAE5pefQEMP25x",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "grok-3-mini-beta_high",
      "score": 0.7462121212121212,
      "bestScore": 0.7550505050505051,
      "standardError": 0.027675521820041432,
      "startedAt": "2025-05-26T18:38:04.231Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJFZMWFVGNi5R5TdZrvQUFw.eval",
      "sourceId": "JFZMWFVGNi5R5TdZrvQUFw",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "DeepSeek-R1",
      "score": 0.6922348484848484,
      "bestScore": 0.7171717171717171,
      "standardError": 0.03070485693321271,
      "startedAt": "2025-05-26T18:38:03.702Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfFGyQR2mfDufvrSV2rzWyV.eval",
      "sourceId": "fFGyQR2mfDufvrSV2rzWyV",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-20250514_59K",
      "score": 0.7781991873045173,
      "bestScore": 0.7918781725888325,
      "standardError": 0.02674821434321096,
      "startedAt": "2025-05-26T18:38:03.090Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJ9wjcQqKJ7hCSYnnoa5E9i.eval",
      "sourceId": "J9wjcQqKJ7hCSYnnoa5E9i",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "grok-3-beta",
      "score": 0.6758207070707071,
      "bestScore": 0.7575757575757576,
      "standardError": 0.026863078746340353,
      "startedAt": "2025-05-26T17:42:41.696Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFQ8Ykq4JWRSXnyF9rbmLc7.eval",
      "sourceId": "FQ8Ykq4JWRSXnyF9rbmLc7",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-20250514_32K",
      "score": 0.7828282828282829,
      "bestScore": 0.7828282828282829,
      "standardError": 0.02937661648494561,
      "startedAt": "2025-05-22T23:06:11.598Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F8hGbWtbpUjAQpebkKQfSi7.eval",
      "sourceId": "8hGbWtbpUjAQpebkKQfSi7",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-20250514_16K",
      "score": 0.7626262626262627,
      "bestScore": 0.7626262626262627,
      "standardError": 0.03031371053819892,
      "startedAt": "2025-05-22T22:54:06.726Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCtxFusPxCraL3EeRJu2c5k.eval",
      "sourceId": "CtxFusPxCraL3EeRJu2c5k",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-20250514_16K",
      "score": 0.7575757575757576,
      "bestScore": 0.7575757575757576,
      "standardError": 0.030532892233932022,
      "startedAt": "2025-05-22T22:54:05.316Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPPLoBnbKQ9CEYZNSk3E7LE.eval",
      "sourceId": "PPLoBnbKQ9CEYZNSk3E7LE",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-sonnet-4-20250514",
      "score": 0.6666666666666666,
      "bestScore": 0.6666666666666666,
      "standardError": 0.033586181457325226,
      "startedAt": "2025-05-22T17:26:47.959Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZF4eejw7J2KrcoNWd6hUbU.eval",
      "sourceId": "ZF4eejw7J2KrcoNWd6hUbU",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-opus-4-20250514",
      "score": 0.6919191919191919,
      "bestScore": 0.6919191919191919,
      "standardError": 0.03289477330098615,
      "startedAt": "2025-05-22T17:26:47.111Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FEnjJ8bGi9vRNNQFzfkLCkQ.eval",
      "sourceId": "EnjJ8bGi9vRNNQFzfkLCkQ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "mistral-medium-2505",
      "score": 0.5953282828282829,
      "bestScore": 0.5953282828282829,
      "standardError": 0.02824697229654645,
      "startedAt": "2025-05-07T17:32:22.166Z",
      "releaseDate": "2025-05-07",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFNMbfvEEm9PRnrJnrqk2PH.eval",
      "sourceId": "FNMbfvEEm9PRnrJnrqk2PH",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o4-mini-2025-04-16_high",
      "score": 0.7960858585858586,
      "bestScore": 0.7960858585858586,
      "standardError": 0.024034764096121958,
      "startedAt": "2025-04-16T18:17:16.139Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9Sgsz5XCjscuZkexf7MX9c.eval",
      "sourceId": "9Sgsz5XCjscuZkexf7MX9c",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o3-2025-04-16_high",
      "score": 0.8181818181818182,
      "bestScore": 0.8181818181818182,
      "standardError": 0.021267123846387202,
      "startedAt": "2025-04-16T18:17:09.093Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FevANZQ9oQTdGYbDERw7VrD.eval",
      "sourceId": "evANZQ9oQTdGYbDERw7VrD",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4.1-mini-2025-04-14",
      "score": 0.6584595959595959,
      "bestScore": 0.6584595959595959,
      "standardError": 0.027051500965295915,
      "startedAt": "2025-04-14T22:49:57.790Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdXHddeGRX7JRVa5UNfad5C.eval",
      "sourceId": "dXHddeGRX7JRVa5UNfad5C",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4.1-nano-2025-04-14",
      "score": 0.48926767676767674,
      "bestScore": 0.48926767676767674,
      "standardError": 0.024849734950435368,
      "startedAt": "2025-04-14T22:49:56.708Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FazWfeFPP6zSocuQX3JYBU5.eval",
      "sourceId": "azWfeFPP6zSocuQX3JYBU5",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4.1-2025-04-14",
      "score": 0.6691919191919192,
      "bestScore": 0.6691919191919192,
      "standardError": 0.02821440093855908,
      "startedAt": "2025-04-14T18:34:33.975Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGj8pd5gU9r4LqwDK8XZ4mL.eval",
      "sourceId": "Gj8pd5gU9r4LqwDK8XZ4mL",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwq-plus",
      "score": 0.6540404040404041,
      "bestScore": 0.6540404040404041,
      "standardError": 0.028469238770829935,
      "startedAt": "2025-04-11T15:34:21.181Z",
      "releaseDate": "2025-04-08",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPtxHF8BH8WuNdNfDnutGBb.eval",
      "sourceId": "PtxHF8BH8WuNdNfDnutGBb",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "grok-3-mini-beta_low",
      "score": 0.7626262626262627,
      "bestScore": 0.7626262626262627,
      "standardError": 0.03031371053819892,
      "startedAt": "2025-04-10T20:30:33.024Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMcxfQdC5aypnD3STwZUBjH.eval",
      "sourceId": "McxfQdC5aypnD3STwZUBjH",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Llama-4-Scout-17B-16E-Instruct",
      "score": 0.5183080808080808,
      "bestScore": 0.5183080808080808,
      "standardError": 0.031549020124994755,
      "startedAt": "2025-04-08T08:34:50.470Z",
      "releaseDate": "2025-04-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMsXAXDEKP3xBKtiA7aTBBG.eval",
      "sourceId": "MsXAXDEKP3xBKtiA7aTBBG",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Llama-4-Maverick-17B-128E-Instruct-FP8",
      "score": 0.6698232323232324,
      "bestScore": 0.6698232323232324,
      "standardError": 0.028244823529728173,
      "startedAt": "2025-04-08T08:34:49.725Z",
      "releaseDate": "2025-04-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLyHy99ubGCqBhoaksQjkiF.eval",
      "sourceId": "LyHy99ubGCqBhoaksQjkiF",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen-turbo-2024-11-01",
      "score": 0.41792929292929293,
      "bestScore": 0.41792929292929293,
      "standardError": 0.02617727508870251,
      "startedAt": "2025-04-07T21:38:51.185Z",
      "releaseDate": "2024-11-01",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9Ke5At9bYuf56SmKJJ7Sz7.eval",
      "sourceId": "9Ke5At9bYuf56SmKJJ7Sz7",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen-plus-2025-01-25",
      "score": 0.4810606060606061,
      "bestScore": 0.4810606060606061,
      "standardError": 0.027144936812277986,
      "startedAt": "2025-04-07T19:28:06.394Z",
      "releaseDate": "2025-01-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQtraMoKiQRRpZc9txpFAYF.eval",
      "sourceId": "QtraMoKiQRRpZc9txpFAYF",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen-max-2025-01-25",
      "score": 0.5612373737373737,
      "bestScore": 0.5612373737373737,
      "standardError": 0.027858831475709448,
      "startedAt": "2025-04-01T19:22:56.011Z",
      "releaseDate": "2025-01-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGBf7rqHkGGzfX87V37F9mH.eval",
      "sourceId": "GBf7rqHkGGzfX87V37F9mH",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "DeepSeek-V3-0324",
      "score": 0.6761363636363636,
      "bestScore": 0.6761363636363636,
      "standardError": 0.026836302525030095,
      "startedAt": "2025-04-01T14:40:59.363Z",
      "releaseDate": "2025-03-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVjfLwKbT6kTprYsMGcz2YT.eval",
      "sourceId": "VjfLwKbT6kTprYsMGcz2YT",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-2.5-pro-exp-03-25",
      "score": 0.8383838383838383,
      "bestScore": 0.8383838383838383,
      "standardError": 0.02622591986362927,
      "startedAt": "2025-03-31T16:06:24.669Z",
      "releaseDate": "2025-03-25",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWKjzEQdfnBHnYeZoKeCWnB.eval",
      "sourceId": "WKjzEQdfnBHnYeZoKeCWnB",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "mistral-small-2503",
      "score": 0.47474747474747475,
      "bestScore": 0.47474747474747475,
      "standardError": 0.027616495140012243,
      "startedAt": "2025-03-18T16:08:06.162Z",
      "releaseDate": "2025-03-17",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FS8bBDAbucsFceirHxeEDvu.eval",
      "sourceId": "S8bBDAbucsFceirHxeEDvu",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-3-5-haiku-20241022",
      "score": 0.3813131313131313,
      "bestScore": 0.3813131313131313,
      "standardError": 0.026351515390511556,
      "startedAt": "2025-03-12T23:51:44.119Z",
      "releaseDate": "2024-10-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJpLxrpvthx7A75sbaHZ9Et.eval",
      "sourceId": "JpLxrpvthx7A75sbaHZ9Et",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-1.5-flash-8b-001",
      "score": 0.32954545454545453,
      "bestScore": 0.32954545454545453,
      "standardError": 0.02416551116211757,
      "startedAt": "2025-03-12T21:23:34.734Z",
      "releaseDate": "2024-10-03",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgFT36W4KU3wggLQ9nNCctg.eval",
      "sourceId": "gFT36W4KU3wggLQ9nNCctg",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "DeepSeek-R1-Distill-Qwen-14B",
      "score": 0.44696969696969696,
      "bestScore": 0.44696969696969696,
      "standardError": 0.02863094678190378,
      "startedAt": "2025-03-10T23:03:48.213Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F8UqaLobhmNJLuPGTTpeaCw.eval",
      "sourceId": "8UqaLobhmNJLuPGTTpeaCw",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-3-7-sonnet-20250219_32K",
      "score": 0.7676767676767676,
      "bestScore": 0.7676767676767676,
      "standardError": 0.030088629490217445,
      "startedAt": "2025-03-10T22:05:35.890Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fiv9yfwZ9vTznxQfvJF6J7p.eval",
      "sourceId": "iv9yfwZ9vTznxQfvJF6J7p",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "DeepSeek-R1-Distill-Llama-70B",
      "score": 0.5574494949494949,
      "bestScore": 0.5574494949494949,
      "standardError": 0.02973465219179067,
      "startedAt": "2025-03-10T20:08:15.034Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Ff6pnAATjxmrFJS353EjQKi.eval",
      "sourceId": "f6pnAATjxmrFJS353EjQKi",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4.5-preview-2025-02-27",
      "score": 0.6868686868686869,
      "bestScore": 0.6868686868686869,
      "standardError": 0.033042050878136546,
      "startedAt": "2025-02-28T09:17:12.859Z",
      "releaseDate": "2025-02-27",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdodibSv7pk8GtqeMy4YLQg.eval",
      "sourceId": "dodibSv7pk8GtqeMy4YLQg",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4-turbo-2024-04-09",
      "score": 0.4659090909090909,
      "bestScore": 0.4659090909090909,
      "standardError": 0.026576883332580552,
      "startedAt": "2025-02-27T14:29:45.537Z",
      "releaseDate": "2024-04-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTaCoN7eyPQEfDWXqs2MpcF.eval",
      "sourceId": "TaCoN7eyPQEfDWXqs2MpcF",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-3-7-sonnet-20250219_16K",
      "score": 0.7676767676767676,
      "bestScore": 0.7676767676767676,
      "standardError": 0.030088629490217445,
      "startedAt": "2025-02-26T16:56:43.203Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVn3Ug88CRbxEaJhLbRzAWD.eval",
      "sourceId": "Vn3Ug88CRbxEaJhLbRzAWD",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "mistral-large-2411",
      "score": 0.5132575757575758,
      "bestScore": 0.5132575757575758,
      "standardError": 0.026887818295509087,
      "startedAt": "2025-02-25T12:31:34.069Z",
      "releaseDate": "2024-11-18",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fn9gpXfUzo8SbWqvrG2q4xM.eval",
      "sourceId": "n9gpXfUzo8SbWqvrG2q4xM",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-3-7-sonnet-20250219",
      "score": 0.6603535353535354,
      "bestScore": 0.6603535353535354,
      "standardError": 0.026703858684060923,
      "startedAt": "2025-02-24T19:14:07.847Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPyNEkFCfMEReodeE2oiPhh.eval",
      "sourceId": "PyNEkFCfMEReodeE2oiPhh",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o1-2024-12-17_high",
      "score": 0.7676767676767676,
      "bestScore": 0.7676767676767676,
      "standardError": 0.030088629490217445,
      "startedAt": "2025-02-13T17:12:11.445Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FNAkCFiFiDSN3NXMj7HMvMS.eval",
      "sourceId": "NAkCFiFiDSN3NXMj7HMvMS",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o1-mini-2024-09-12_high",
      "score": 0.6237373737373737,
      "bestScore": 0.6237373737373737,
      "standardError": 0.027427723511230753,
      "startedAt": "2025-02-13T17:07:50.981Z",
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FX2gKewiGPZx5DLFEddUUqW.eval",
      "sourceId": "X2gKewiGPZx5DLFEddUUqW",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o3-mini-2025-01-31_high",
      "score": 0.7702020202020202,
      "bestScore": 0.7702020202020202,
      "standardError": 0.025855332698878485,
      "startedAt": "2025-02-13T17:07:50.336Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMzsTvCEEfEpzn4WvxYdJkX.eval",
      "sourceId": "MzsTvCEEfEpzn4WvxYdJkX",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-2.0-pro-exp-02-05",
      "score": 0.6565656565656566,
      "bestScore": 0.6565656565656566,
      "standardError": 0.03383201223244441,
      "startedAt": "2025-02-07T07:28:03.988Z",
      "releaseDate": "2025-02-05",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fg5YE7G6DXb45EH4HxEdiht.eval",
      "sourceId": "g5YE7G6DXb45EH4HxEdiht",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-2.0-flash-thinking-exp-01-21",
      "score": 0.5707070707070707,
      "bestScore": 0.5707070707070707,
      "standardError": 0.03526552724601199,
      "startedAt": "2025-02-06T22:40:04.398Z",
      "releaseDate": "2025-01-21",
      "organization": "Google DeepMind,Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fag9w2w4xa47ghZaWt5oZey.eval",
      "sourceId": "ag9w2w4xa47ghZaWt5oZey",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-2.0-flash-001",
      "score": 0.6414141414141414,
      "bestScore": 0.6414141414141414,
      "standardError": 0.027395618179027574,
      "startedAt": "2025-02-06T15:51:21.852Z",
      "releaseDate": "2025-02-05",
      "organization": "Google DeepMind,Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfbszwTApm3f28z5VA6oXBK.eval",
      "sourceId": "fbszwTApm3f28z5VA6oXBK",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4o-2024-11-20",
      "score": 0.4788510101010101,
      "bestScore": 0.4788510101010101,
      "standardError": 0.026336972178726107,
      "startedAt": "2025-02-05T08:31:15.914Z",
      "releaseDate": "2024-11-20",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgULKGqAe9nG9syxmYMtkhi.eval",
      "sourceId": "gULKGqAe9nG9syxmYMtkhi",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "phi-4",
      "score": 0.5606060606060606,
      "bestScore": 0.5606060606060606,
      "standardError": 0.02590274664132679,
      "startedAt": "2025-01-31T23:03:31.781Z",
      "releaseDate": "2024-12-12",
      "organization": "Microsoft Research",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRAwuPRwJedy7P2KeQhEmr8.eval",
      "sourceId": "RAwuPRwJedy7P2KeQhEmr8",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Phi-3-medium-128k-instruct",
      "score": 0.2758838383838384,
      "bestScore": 0.2758838383838384,
      "standardError": 0.016117685500292727,
      "startedAt": "2025-01-31T22:41:25.625Z",
      "releaseDate": "2024-04-23",
      "organization": "Microsoft",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbDS2AGEf3AywbVcdDyAKZK.eval",
      "sourceId": "bDS2AGEf3AywbVcdDyAKZK",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o3-mini-2025-01-31_medium",
      "score": 0.742739898989899,
      "bestScore": 0.742739898989899,
      "standardError": 0.025411943362857046,
      "startedAt": "2025-01-31T19:21:57.338Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FehvA5nisC7GMbgbyy3Z4Et.eval",
      "sourceId": "ehvA5nisC7GMbgbyy3Z4Et",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "mistral-small-2501",
      "score": 0.4529671717171717,
      "bestScore": 0.4529671717171717,
      "standardError": 0.024794688677057853,
      "startedAt": "2025-01-30T21:10:36.084Z",
      "releaseDate": "2025-01-25",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FoDq358GfyV5Au65scERdVo.eval",
      "sourceId": "oDq358GfyV5Au65scERdVo",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen2.5-32b-instruct",
      "score": 0.46085858585858586,
      "bestScore": 0.46085858585858586,
      "standardError": 0.02595783004579065,
      "startedAt": "2025-01-30T18:31:31.252Z",
      "releaseDate": "2024-09-17",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fd6JihAFVqUy3u9suNt4P3L.eval",
      "sourceId": "d6JihAFVqUy3u9suNt4P3L",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-1.5-flash-001",
      "score": 0.40372474747474746,
      "bestScore": 0.40372474747474746,
      "standardError": 0.023852475604024478,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-05-23",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FHc2ycMP6DyCZxCPjKbZFQU.eval",
      "sourceId": "Hc2ycMP6DyCZxCPjKbZFQU",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-2.0",
      "score": 0.3465909090909091,
      "bestScore": 0.3465909090909091,
      "standardError": 0.025076255983551475,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-07-11",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmTa4f52zsEuxdrPy7WfQvX.eval",
      "sourceId": "mTa4f52zsEuxdrPy7WfQvX",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-3-opus-20240229",
      "score": 0.4715909090909091,
      "bestScore": 0.4715909090909091,
      "standardError": 0.026395393956533515,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-02-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhTnDa4gn8RYduCEYqoNywe.eval",
      "sourceId": "hTnDa4gn8RYduCEYqoNywe",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o1-mini-2024-09-12_medium",
      "score": 0.5950126262626263,
      "bestScore": 0.5950126262626263,
      "standardError": 0.02795124238637276,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F4NiUpCrLfz3oFaerkrMeBt.eval",
      "sourceId": "4NiUpCrLfz3oFaerkrMeBt",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4o-2024-08-06",
      "score": 0.49210858585858586,
      "bestScore": 0.49210858585858586,
      "standardError": 0.025999270277271707,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-08-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfairJpL9VekDBKxYHG9ZoG.eval",
      "sourceId": "fairJpL9VekDBKxYHG9ZoG",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o1-preview-2024-09-12",
      "score": 0.5031565656565656,
      "bestScore": 0.5031565656565656,
      "standardError": 0.028257821072909393,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7CkADTsoNbDd88uqYbqfwj.eval",
      "sourceId": "7CkADTsoNbDd88uqYbqfwj",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-3-sonnet-20240229",
      "score": 0.4059343434343434,
      "bestScore": 0.4059343434343434,
      "standardError": 0.023920223491748827,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-02-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQLejPZXhjX6SQCQgzjNvCG.eval",
      "sourceId": "QLejPZXhjX6SQCQgzjNvCG",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-3-5-sonnet-20241022",
      "score": 0.553030303030303,
      "bestScore": 0.553030303030303,
      "standardError": 0.028208090147078847,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-10-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJCCVcVLz3J4TQXSU67hMJZ.eval",
      "sourceId": "JCCVcVLz3J4TQXSU67hMJZ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4-0125-preview",
      "score": 0.42266414141414144,
      "bestScore": 0.42266414141414144,
      "standardError": 0.026795949166711065,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-01-25",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F4j3WRhhruXrfaq9Jki3NDc.eval",
      "sourceId": "4j3WRhhruXrfaq9Jki3NDc",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-2.1",
      "score": 0.32954545454545453,
      "bestScore": 0.32954545454545453,
      "standardError": 0.02324885083265289,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-11-21",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FeMAisrJd7fAUHPfcPWXeXo.eval",
      "sourceId": "eMAisrJd7fAUHPfcPWXeXo",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemma-2-27b-it",
      "score": 0.3648989898989899,
      "bestScore": 0.3648989898989899,
      "standardError": 0.023897884428691632,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-06-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZUJxsJDpHnLXxNiDcvj22W.eval",
      "sourceId": "ZUJxsJDpHnLXxNiDcvj22W",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-3-haiku-20240307",
      "score": 0.3630050505050505,
      "bestScore": 0.3630050505050505,
      "standardError": 0.024180910672074975,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-03-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FT3FWdCCWBryVjxwTcoY5bK.eval",
      "sourceId": "T3FWdCCWBryVjxwTcoY5bK",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "claude-3-5-sonnet-20240620",
      "score": 0.5404040404040404,
      "bestScore": 0.5404040404040404,
      "standardError": 0.027528555453348653,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-06-20",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F78Wpvp8WDByAcQRBas72cs.eval",
      "sourceId": "78Wpvp8WDByAcQRBas72cs",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4o-2024-05-13",
      "score": 0.4889520202020202,
      "bestScore": 0.4889520202020202,
      "standardError": 0.026254109873371994,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-05-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FY59q8WbEDp2HH8oovyJCj9.eval",
      "sourceId": "Y59q8WbEDp2HH8oovyJCj9",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-1.5-flash-002",
      "score": 0.47316919191919193,
      "bestScore": 0.47316919191919193,
      "standardError": 0.027737408277040958,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPCLvrs4TEBtqQqcYeU8pAG.eval",
      "sourceId": "PCLvrs4TEBtqQqcYeU8pAG",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4-0613",
      "score": 0.30650252525252525,
      "bestScore": 0.30650252525252525,
      "standardError": 0.02337322948270623,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-06-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Ffu7pWqbSC4MUBdwv3GFNa5.eval",
      "sourceId": "fu7pWqbSC4MUBdwv3GFNa5",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemma-2-9b-it",
      "score": 0.2746212121212121,
      "bestScore": 0.2746212121212121,
      "standardError": 0.023032467954778795,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-06-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfQjoHKyUEPAuwdrBGfL8SG.eval",
      "sourceId": "fQjoHKyUEPAuwdrBGfL8SG",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-3.5-turbo-1106",
      "score": 0.2803030303030303,
      "bestScore": 0.2803030303030303,
      "standardError": 0.018720372561004013,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-11-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FenhRcQmfN6RFB2zu7R3Uxr.eval",
      "sourceId": "enhRcQmfN6RFB2zu7R3Uxr",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-1.5-pro-001",
      "score": 0.4586489898989899,
      "bestScore": 0.4586489898989899,
      "standardError": 0.02617295639558329,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-05-14",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPKdqpKZ9tqDWyWJ7GksBWz.eval",
      "sourceId": "PKdqpKZ9tqDWyWJ7GksBWz",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4-1106-preview",
      "score": 0.4236111111111111,
      "bestScore": 0.4236111111111111,
      "standardError": 0.024076142618301676,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-11-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FAWewipyfSQb8hXtS8d4fNL.eval",
      "sourceId": "AWewipyfSQb8hXtS8d4fNL",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-4o-mini-2024-07-18",
      "score": 0.37720959595959597,
      "bestScore": 0.37720959595959597,
      "standardError": 0.024068925424620136,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-07-18",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FL9yfwZC3snsKZshBXTaXjc.eval",
      "sourceId": "L9yfwZC3snsKZshBXTaXjc",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-1.5-pro-002",
      "score": 0.5722853535353535,
      "bestScore": 0.5722853535353535,
      "standardError": 0.027959709636504244,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9TrLEEw8TvsyugyCjwvUvD.eval",
      "sourceId": "9TrLEEw8TvsyugyCjwvUvD",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gpt-3.5-turbo-0125",
      "score": 0.27178030303030304,
      "bestScore": 0.27178030303030304,
      "standardError": 0.01953579968788737,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-01-25",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FeZaQhpbhbzbo3sbKGS5Efa.eval",
      "sourceId": "eZaQhpbhbzbo3sbKGS5Efa",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Llama-3.1-70B-Instruct",
      "score": 0.44191919191919193,
      "bestScore": 0.44191919191919193,
      "standardError": 0.024522863472149384,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMEwMcmSV32JHrsNoAq4BFL.eval",
      "sourceId": "MEwMcmSV32JHrsNoAq4BFL",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Llama-3.1-405B-Instruct",
      "score": 0.5091540404040404,
      "bestScore": 0.5091540404040404,
      "standardError": 0.025861992702770648,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FS5QYXSvQBRSbUbXSnAGbMm.eval",
      "sourceId": "S5QYXSvQBRSbUbXSnAGbMm",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Yi-1.5-34B-Chat",
      "score": 0.319760101010101,
      "bestScore": 0.319760101010101,
      "standardError": 0.020005291661900883,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-05-13",
      "organization": "01.AI",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJwi8JTeeTxJuJY8qidxngw.eval",
      "sourceId": "Jwi8JTeeTxJuJY8qidxngw",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Yi-34B-Chat",
      "score": 0.14741161616161616,
      "bestScore": 0.14741161616161616,
      "standardError": 0.011372823721448395,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-11-22",
      "organization": "01.AI",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWxMpEBcTf8qPFPz4BrF86B.eval",
      "sourceId": "WxMpEBcTf8qPFPz4BrF86B",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "grok-2-1212",
      "score": 0.5378787878787878,
      "bestScore": 0.5378787878787878,
      "standardError": 0.027066342316607196,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-12-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F83DRJD4FvjwqU9eStRaNiU.eval",
      "sourceId": "83DRJD4FvjwqU9eStRaNiU",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen2-72b-instruct",
      "score": 0.4078282828282828,
      "bestScore": 0.4078282828282828,
      "standardError": 0.02574288051254005,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-06-07",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPex6SpYJX8ZPySncfNNy7H.eval",
      "sourceId": "Pex6SpYJX8ZPySncfNNy7H",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen1.5-72b-chat",
      "score": 0.2881944444444444,
      "bestScore": 0.2881944444444444,
      "standardError": 0.01993691228122493,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-02-04",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FNeqeQSxHmdUNLk9Ytpo6qo.eval",
      "sourceId": "NeqeQSxHmdUNLk9Ytpo6qo",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen1.5-32b-chat",
      "score": 0.307449494949495,
      "bestScore": 0.307449494949495,
      "standardError": 0.020200517973424446,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-04-03",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9K6c9b6v9zQrQYtw827BDn.eval",
      "sourceId": "9K6c9b6v9zQrQYtw827BDn",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Hermes-2-Theta-Llama-3-70B",
      "score": 0.3746843434343434,
      "bestScore": 0.3746843434343434,
      "standardError": 0.024803009993948306,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-06-20",
      "organization": "Nous Research,Arcee AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDHdeZNynu4VtRp2VGkCGvd.eval",
      "sourceId": "DHdeZNynu4VtRp2VGkCGvd",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Llama-2-70b-chat-hf",
      "score": 0.26325757575757575,
      "bestScore": 0.26325757575757575,
      "standardError": 0.019901808558284103,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-07-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FevPjJw9SKqq3TieSdxrSQn.eval",
      "sourceId": "evPjJw9SKqq3TieSdxrSQn",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Meta-Llama-3-70B-Instruct",
      "score": 0.40561868686868685,
      "bestScore": 0.40561868686868685,
      "standardError": 0.026224849492112068,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-04-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjrvutCoFFG4KxYtU7REdpe.eval",
      "sourceId": "jrvutCoFFG4KxYtU7REdpe",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Meta-Llama-3-8B-Instruct",
      "score": 0.26073232323232326,
      "bestScore": 0.26073232323232326,
      "standardError": 0.01703043361773576,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-04-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FiT3BBFWjsZk2rKHKeJDsM8.eval",
      "sourceId": "iT3BBFWjsZk2rKHKeJDsM8",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Mistral-7B-Instruct-v0.3",
      "score": 0.1518308080808081,
      "bestScore": 0.1518308080808081,
      "standardError": 0.010148891037227303,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-05-27",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYbQkMPC2GtNLhxqTEhhVsa.eval",
      "sourceId": "YbQkMPC2GtNLhxqTEhhVsa",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "deepseek-llm-67b-chat",
      "score": 0.24621212121212122,
      "bestScore": 0.24621212121212122,
      "standardError": 0.013623351950761947,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-11-29",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F82RTgqgiC2UuM5M37L3WWM.eval",
      "sourceId": "82RTgqgiC2UuM5M37L3WWM",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Mixtral-8x7B-Instruct-v0.1",
      "score": 0.3058712121212121,
      "bestScore": 0.3058712121212121,
      "standardError": 0.01937059603941351,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-12-11",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSn2qbB2HiySe7NcG28jnyM.eval",
      "sourceId": "Sn2qbB2HiySe7NcG28jnyM",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "qwen2.5-72b-instruct",
      "score": 0.4914772727272727,
      "bestScore": 0.4914772727272727,
      "standardError": 0.026947767404178667,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWtALgW6VSFdihDZWXRm2AZ.eval",
      "sourceId": "WtALgW6VSFdihDZWXRm2AZ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "WizardLM-2-8x22B",
      "score": 0.43434343434343436,
      "bestScore": 0.43434343434343436,
      "standardError": 0.023232093333444247,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-04-15",
      "organization": "Microsoft",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMQJA3RUjsQzSubGJiukZ7A.eval",
      "sourceId": "MQJA3RUjsQzSubGJiukZ7A",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "dbrx-instruct",
      "score": 0.32891414141414144,
      "bestScore": 0.32891414141414144,
      "standardError": 0.03256944333564935,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-03-27",
      "organization": "Databricks",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTTa3hu7PgRxHgizVreZsdJ.eval",
      "sourceId": "TTa3hu7PgRxHgizVreZsdJ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "gemini-1.0-pro-001",
      "score": 0.33964646464646464,
      "bestScore": 0.33964646464646464,
      "standardError": 0.020110933405176737,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-12-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FibpS8bs4z3wD3Hb6F3kpdV.eval",
      "sourceId": "ibpS8bs4z3wD3Hb6F3kpdV",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "ministral-3b-2410",
      "score": 0.25252525252525254,
      "bestScore": 0.25252525252525254,
      "standardError": 0.019305614722797293,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-10-16",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbJV4D8SEo9zMJPEYTTXJNZ.eval",
      "sourceId": "bJV4D8SEo9zMJPEYTTXJNZ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "mistral-large-2402",
      "score": 0.38762626262626265,
      "bestScore": 0.38762626262626265,
      "standardError": 0.024776598535417284,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-02-26",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdDeREdDJ5bVyci3APZxzWb.eval",
      "sourceId": "dDeREdDJ5bVyci3APZxzWb",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "open-mixtral-8x22b",
      "score": 0.34059343434343436,
      "bestScore": 0.34059343434343436,
      "standardError": 0.02068134779158647,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-04-17",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FifAM3DJBdm5eU9Y2DjaEvW.eval",
      "sourceId": "ifAM3DJBdm5eU9Y2DjaEvW",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "ministral-8b-2410",
      "score": 0.27146464646464646,
      "bestScore": 0.27146464646464646,
      "standardError": 0.020863706671553732,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-10-16",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDGEzZA6nvTne4MEGSuo3Tm.eval",
      "sourceId": "DGEzZA6nvTne4MEGSuo3Tm",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "mistral-large-2407",
      "score": 0.49021464646464646,
      "bestScore": 0.49021464646464646,
      "standardError": 0.025918840988656967,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-07-24",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZQPNnicTuhuscJvni85PFY.eval",
      "sourceId": "ZQPNnicTuhuscJvni85PFY",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "open-mixtral-8x7b",
      "score": 0.29829545454545453,
      "bestScore": 0.29829545454545453,
      "standardError": 0.019146998037923697,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-12-11",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FY6KPSujFepdwqinKLf9mDF.eval",
      "sourceId": "Y6KPSujFepdwqinKLf9mDF",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "open-mistral-7b",
      "score": 0.13226010101010102,
      "bestScore": 0.13226010101010102,
      "standardError": 0.009888086213040367,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-09-27",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F5XuVuHeUDekHqjUy2P8gww.eval",
      "sourceId": "5XuVuHeUDekHqjUy2P8gww",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "open-mistral-nemo-2407",
      "score": 0.2989267676767677,
      "bestScore": 0.2989267676767677,
      "standardError": 0.02096191098761666,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-07-18",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXCSWtbDquA7FUiWfd7kh9q.eval",
      "sourceId": "XCSWtbDquA7FUiWfd7kh9q",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Llama-3.2-90B-Vision-Instruct",
      "score": 0.41035353535353536,
      "bestScore": 0.41035353535353536,
      "standardError": 0.022948341091025694,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-24",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9SSHEP95FYVZvHDdoHSRy3.eval",
      "sourceId": "9SSHEP95FYVZvHDdoHSRy3",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Llama-3.1-Tulu-3-70B-DPO",
      "score": 0.46275252525252525,
      "bestScore": 0.46275252525252525,
      "standardError": 0.02405836362093961,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-11-21",
      "organization": "Allen Institute for AI,University of Washington",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F5JgivNDZp2D7Lc7cpDhS28.eval",
      "sourceId": "5JgivNDZp2D7Lc7cpDhS28",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Eurus-2-7B-PRIME",
      "score": 0.3390151515151515,
      "bestScore": 0.3390151515151515,
      "standardError": 0.02107317439394748,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-12-31",
      "organization": "Tsinghua University,University of Illinois Urbana-Champaign (UIUC),Shanghai AI Lab,Peking University,Shanghai Jiao Tong University,CUHK Shenzhen Research Institute",
      "country": "United States of America,China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRJm4u2vXF6PUmThJGzAfHE.eval",
      "sourceId": "RJm4u2vXF6PUmThJGzAfHE",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "Llama-3.3-70B-Instruct",
      "score": 0.4744318181818182,
      "bestScore": 0.4744318181818182,
      "standardError": 0.028523027562061026,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-12-06",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTxXS78Wg2pSCDbpmeJuzQn.eval",
      "sourceId": "TxXS78Wg2pSCDbpmeJuzQn",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "o1-2024-12-17_medium",
      "score": 0.7575757575757576,
      "bestScore": 0.7575757575757576,
      "standardError": 0.030532892233932022,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FUczhz7MKLstjLkSbbGuwdN.eval",
      "sourceId": "Uczhz7MKLstjLkSbbGuwdN",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "DeepSeek-V3",
      "score": 0.5653409090909091,
      "bestScore": 0.5653409090909091,
      "standardError": 0.02763990029950893,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-12-26",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVwo3nMA8g2gGBCyTdKFMQJ.eval",
      "sourceId": "Vwo3nMA8g2gGBCyTdKFMQJ",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "GPQA diamond",
      "model": "grok-4-0709",
      "score": 0.87,
      "bestScore": 0.87,
      "standardError": 0.02,
      "startedAt": null,
      "releaseDate": "2025-07-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "A85Zfq2qguE4X9xXBweBHP",
      "sourceFile": "gpqa_diamond.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-8_max",
      "score": 0.4724137931034483,
      "bestScore": 0.4724137931034483,
      "standardError": 0.029366965952163048,
      "startedAt": "2026-06-08T00:20:40.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6dUEQ3YfeMZzSm9ytgiuDi",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "qwen3.6-max-preview",
      "score": 0.23103448275862068,
      "bestScore": 0.23103448275862068,
      "standardError": 0.02479378972186167,
      "startedAt": "2026-05-27T22:28:21.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "T6yJ47dEheZTjRVDS6Bgzt",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gemini-3.5-flash_high",
      "score": 0.3896551724137931,
      "bestScore": 0.3896551724137931,
      "standardError": 0.028686590614000477,
      "startedAt": "2026-05-22T16:42:34.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Ectc9Fcn4xSKZFQwj9Xn7R",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "qwen3.5-plus",
      "score": 0.2103448275862069,
      "bestScore": 0.2103448275862069,
      "standardError": 0.02397373715108675,
      "startedAt": "2026-05-14T12:48:49.000Z",
      "releaseDate": "2026-02-16",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "Kbc5vW47MpsnYaqeBuoJoQ",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "qwen3.6-plus",
      "score": 0.2620689655172414,
      "bestScore": 0.2620689655172414,
      "standardError": 0.025868232968648857,
      "startedAt": "2026-05-12T16:08:00.000Z",
      "releaseDate": "2026-03-31",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "Vxki26PKxMtCWsahLMmBHB",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "qwen3.6-flash",
      "score": 0.10344827586206896,
      "bestScore": 0.10344827586206896,
      "standardError": 0.01791432224407271,
      "startedAt": "2026-05-12T15:03:26.000Z",
      "releaseDate": "2026-04-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "n4BkffZyi6KAx8HajDU4ZK",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "qwen3.5-flash",
      "score": 0.06206896551724138,
      "bestScore": 0.06206896551724138,
      "standardError": 0.014192986356621421,
      "startedAt": "2026-05-12T12:27:22.000Z",
      "releaseDate": "2026-02-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "VHC8c6kTi5zjkHU67QSkZW",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "glm-5.1",
      "score": 0.33448275862068966,
      "bestScore": 0.33448275862068966,
      "standardError": 0.027753489933368777,
      "startedAt": "2026-05-11T13:39:21.000Z",
      "releaseDate": "2026-04-07",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "4DNcqcs4R97ZWNTMDrXSZN",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "kimi-k2.6",
      "score": 0.3896551724137931,
      "bestScore": 0.3896551724137931,
      "standardError": 0.028686590614000477,
      "startedAt": "2026-05-07T21:21:29.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "4StdTsQNqgsKc8RFdauiF2",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.5-pro-pre-release_xhigh",
      "score": 0.51,
      "bestScore": 0.51,
      "standardError": 0.029,
      "startedAt": "2026-04-23T20:51:52.501Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gpt-5-5-pro-pre-release-xhigh-t13",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.5-pro-pre-release_high",
      "score": 0.524,
      "bestScore": 0.524,
      "standardError": 0.029,
      "startedAt": "2026-04-23T20:51:39.890Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gpt-5-5-pro-pre-release-high-t13",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.5-pre-release_xhigh",
      "score": 0.517,
      "bestScore": 0.517,
      "standardError": 0.029,
      "startedAt": "2026-04-23T20:51:32.704Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gpt-5-5-pre-release-xhigh-t13",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-7_xhigh",
      "score": 0.4379310344827586,
      "bestScore": 0.4379310344827586,
      "standardError": 0.02918426387376174,
      "startedAt": "2026-04-17T11:24:50.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "VFJMk4X6U8Yx5FwtQh4jbk",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.4-mini-2026-03-17_high",
      "score": 0.2828,
      "bestScore": 0.2828,
      "standardError": 0.02065,
      "startedAt": "2026-04-15T13:54:39.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "7oFhmdG6DwLVLAenx8Wu4z",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.4-nano-2026-03-17_high",
      "score": 0.2586,
      "bestScore": 0.2586,
      "standardError": 0.0258,
      "startedAt": "2026-04-15T09:11:20.668Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "SZpkpKjZLWJvmXWPQsMpbv",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "muse-spark",
      "score": 0.39,
      "bestScore": 0.39,
      "standardError": 0.029,
      "startedAt": "2026-04-08T16:30:10.530Z",
      "releaseDate": "2026-04-08",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "muse-spark-pre-release-tiers-1-3",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.4-2026-03-05_xhigh",
      "score": 0.476,
      "bestScore": 0.476,
      "standardError": 0.029,
      "startedAt": "2026-03-06T03:53:08.691Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "pre-release-gpt-5-4-xhigh-fm-t13",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.4-pro-2026-03-05_xhigh",
      "score": 0.5,
      "bestScore": 0.5,
      "standardError": 0.029,
      "startedAt": "2026-03-06T03:47:23.261Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "pre-release-gpt-5-4-pro-fm-t13",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-sonnet-4-6_16K",
      "score": 0.324,
      "bestScore": 0.324,
      "standardError": 0.028121636040639858,
      "startedAt": "2026-02-20T14:53:58.744Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ewFjUhQhYidjkErDdLvcQL",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "glm-5",
      "score": 0.16433566433566432,
      "bestScore": 0.16433566433566432,
      "standardError": 0.021951258107494095,
      "startedAt": "2026-02-19T19:17:36.176Z",
      "releaseDate": "2026-02-11",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "MEXifXdpy4M5JH6EoyF35G",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gemini-3.1-pro-preview",
      "score": 0.369,
      "bestScore": 0.369,
      "standardError": 0.028,
      "startedAt": "2026-02-19T18:54:42.570Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "CZ5mNHNivNueD6boJnxq6H",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-6_max",
      "score": 0.407,
      "bestScore": 0.407,
      "standardError": 0.029,
      "startedAt": "2026-02-12T17:17:42.489Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Y4von2jMmwvEK5MuNw9pSt",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-6_64K",
      "score": 0.39655172413793105,
      "bestScore": 0.39655172413793105,
      "standardError": 0.028775377199008325,
      "startedAt": "2026-02-06T15:39:35.993Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "QSNvJqsvYdNbb85DyL9AXn",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-6_32K",
      "score": 0.4,
      "bestScore": 0.4,
      "standardError": 0.028817526385684453,
      "startedAt": "2026-02-06T12:46:04.911Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "b2iuVcuCPEau2ryhBsjhS9",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-6",
      "score": 0.38275862068965516,
      "bestScore": 0.38275862068965516,
      "standardError": 0.02859177289764447,
      "startedAt": "2026-02-05T18:49:51.181Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "nFxCq7MqCfKjhxKr7hT9qr",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "fireworks/kimi-k2p5",
      "score": 0.279,
      "bestScore": 0.279,
      "standardError": 0.026,
      "startedAt": "2026-02-03T10:31:18.169Z",
      "releaseDate": "2026-01-27",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "manual_kimi-k2.5_fmprivate_frankenstein",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "zai-org/GLM-4.7",
      "score": 0.024390243902439025,
      "bestScore": 0.024390243902439025,
      "standardError": 0.009121434048862995,
      "startedAt": "2026-01-30T03:56:27.987Z",
      "releaseDate": "2025-12-22",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "GJNr6pF9QUonmhqw5dXWpd",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "fireworks/deepseek-v3p2",
      "score": 0.221,
      "bestScore": 0.221,
      "standardError": 0.041,
      "startedAt": "2025-12-22T16:13:27.305Z",
      "releaseDate": "2025-12-01",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "manual_deepseek-v3.2_frontiermath_tiers_13_private_frankenstein_run_incomplete",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gemini-2.5-flash",
      "score": 0.04844290657439446,
      "bestScore": 0.04844290657439446,
      "standardError": 0.012651332738454618,
      "startedAt": "2025-12-18T18:23:15.700Z",
      "releaseDate": "2025-06-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bTbDNLFqsUogG3fejZEAZc",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gemini-3-flash-preview",
      "score": 0.356401384083045,
      "bestScore": 0.356401384083045,
      "standardError": 0.028221558784583784,
      "startedAt": "2025-12-17T22:40:29.699Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "R7dSNsq2ciVRVHB7nir69v",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.2-2025-12-11_xhigh",
      "score": 0.407,
      "bestScore": 0.407,
      "standardError": 0.029,
      "startedAt": "2025-12-13T13:45:00.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hYCViU82i8tejydKA4Pcuu",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.2-2025-12-11_high",
      "score": 0.40344827586206894,
      "bestScore": 0.40344827586206894,
      "standardError": 0.028858188319787523,
      "startedAt": "2025-12-11T20:49:17.862Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dcQtgoUnQbkqTT6Cueo8Re",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.2-2025-12-11_medium",
      "score": 0.3689655172413793,
      "bestScore": 0.3689655172413793,
      "standardError": 0.02838379845926905,
      "startedAt": "2025-12-11T20:49:17.514Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "BWGVnFerZCyKTuc86w7c5q",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.2-2025-12-11_low",
      "score": 0.2655172413793103,
      "bestScore": 0.2655172413793103,
      "standardError": 0.02597695517914575,
      "startedAt": "2025-12-11T20:24:40.964Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "NghuUppaqcCUtT8BqRgUL6",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "Qwen/Qwen3-235B-A22B-Thinking-2507",
      "score": 0.08480565371024736,
      "bestScore": 0.08480565371024736,
      "standardError": 0.016589928458951776,
      "startedAt": "2025-12-11T17:22:27.962Z",
      "releaseDate": "2025-07-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "TLsGMRt4tYtWnheLm6jA5s",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "zai-org/GLM-4.6",
      "score": 0.03819444444444445,
      "bestScore": 0.03819444444444445,
      "standardError": 0.011313651347924142,
      "startedAt": "2025-12-08T20:08:02.803Z",
      "releaseDate": "2025-09-30",
      "organization": "Z.ai (Zhipu AI),Tsinghua University",
      "country": "China",
      "logViewer": null,
      "sourceId": "MPJBHaxt8mXcKBWa94rsXj",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "moonshotai/Kimi-K2-Thinking",
      "score": 0.21403508771929824,
      "bestScore": 0.21403508771929824,
      "standardError": 0.02433800055326239,
      "startedAt": "2025-12-05T16:22:32.448Z",
      "releaseDate": "2025-11-06",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "FmXuaTot9QsNdFYwwFHuEG",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.1-2025-11-13_none",
      "score": 0.02097902097902098,
      "bestScore": 0.02097902097902098,
      "standardError": 0.008489188512823553,
      "startedAt": "2025-11-25T22:22:28.656Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XNbh4FSUJeFAe2auBg5Bdi",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.1-2025-11-13_low",
      "score": 0.17301038062283736,
      "bestScore": 0.17301038062283736,
      "standardError": 0.022288963139763994,
      "startedAt": "2025-11-25T15:12:29.299Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "7o6HCQaKUrxWjrf2vhH7hV",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-5-20251101",
      "score": 0.20689655172413793,
      "bestScore": 0.20689655172413793,
      "standardError": 0.02382827611454509,
      "startedAt": "2025-11-25T04:02:38.737Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9nctB7cH6kcLT24UbScEMa",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-5-20251101_32K",
      "score": 0.20689655172413793,
      "bestScore": 0.20689655172413793,
      "standardError": 0.02382827611454509,
      "startedAt": "2025-11-25T01:11:04.094Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "DVjCqcKsfQ4kJbwucCiuH7",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-5-20251101_16K",
      "score": 0.20344827586206896,
      "bestScore": 0.20344827586206896,
      "standardError": 0.023680184210831092,
      "startedAt": "2025-11-25T01:10:48.342Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Xvp9LtJuwPtSab63ZWwkTC",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gemini-2.5-pro",
      "score": 0.1413793103448276,
      "bestScore": 0.1413793103448276,
      "standardError": 0.020494847099131453,
      "startedAt": "2025-11-24T02:37:34.534Z",
      "releaseDate": "2025-06-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "cA9LKKKzDVxiLRcegKSb2b",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gemini-3-pro-preview",
      "score": 0.376,
      "bestScore": 0.376,
      "standardError": 0.0284,
      "startedAt": "2025-11-21T12:00:00.000Z",
      "releaseDate": "2025-11-18",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "manual_gemini_3_fmprivate_frankenstein",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.1-2025-11-13_medium",
      "score": 0.2689655172413793,
      "bestScore": 0.2689655172413793,
      "standardError": 0.026083646905766342,
      "startedAt": "2025-11-17T21:28:57.104Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "7c2hHEgF9JomG9X7BLP7By",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "o3-2025-04-16_low",
      "score": 0.09722222222222222,
      "bestScore": 0.09722222222222222,
      "standardError": 0.017487682248157596,
      "startedAt": "2025-11-17T01:17:15.370Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "DKLrszPeTvFUGq39AaTniX",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "o3-2025-04-16_high",
      "score": 0.18685121107266436,
      "bestScore": 0.18685121107266436,
      "standardError": 0.022968716344899227,
      "startedAt": "2025-11-16T20:35:07.675Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XQ46BgPhHaYZsrPQuTvz8J",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "o3-2025-04-16_medium",
      "score": 0.16896551724137931,
      "bestScore": 0.16896551724137931,
      "standardError": 0.022042438716799215,
      "startedAt": "2025-11-16T20:35:03.993Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XwNqEvxaaCQ8M2NpKJGoZG",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-sonnet-4-5-20250929",
      "score": 0.09310344827586207,
      "bestScore": 0.09310344827586207,
      "standardError": 0.017092785280147835,
      "startedAt": "2025-11-16T12:36:38.474Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "n26C8XHigC9Xfsdb24JZtd",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-sonnet-4-5-20250929_32K",
      "score": 0.1522491349480969,
      "bestScore": 0.1522491349480969,
      "standardError": 0.021169728769293714,
      "startedAt": "2025-11-16T12:36:27.931Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LUpSd7DzWf3Fw4iwXb3MPV",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "o4-mini-2025-04-16_low",
      "score": 0.10689655172413794,
      "bestScore": 0.10689655172413794,
      "standardError": 0.018175392977158443,
      "startedAt": "2025-11-16T03:21:42.143Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "82dL9ukqav96KZ8aHfqPpq",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "o3-mini-2025-01-31_high",
      "score": 0.12413793103448276,
      "bestScore": 0.12413793103448276,
      "standardError": 0.019396402576684655,
      "startedAt": "2025-11-16T01:50:26.218Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "W2YTmg94XqVbH7p86TAqJT",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5.1-2025-11-13_high",
      "score": 0.3103448275862069,
      "bestScore": 0.3103448275862069,
      "standardError": 0.02721380905679257,
      "startedAt": "2025-11-13T20:36:07.951Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "nveMCdbvDijDTBN9JPWNve",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.32413793103448274,
      "bestScore": 0.32413793103448274,
      "standardError": 0.027532461664839192,
      "startedAt": "2025-11-13T19:38:13.991Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "JwfCZuni4TRRnA77UWwpyg",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-sonnet-4-5-20250929_59K",
      "score": 0.13494809688581316,
      "bestScore": 0.13494809688581316,
      "standardError": 0.020132987375676092,
      "startedAt": "2025-11-13T15:54:58.691Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "3GzYKQW2K4iijm6a73rVeh",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5-mini-2025-08-07_high",
      "score": 0.27241379310344827,
      "bestScore": 0.27241379310344827,
      "standardError": 0.02618833296520288,
      "startedAt": "2025-11-13T15:54:56.299Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "DkMXbLvTgWJDWnUQVuBA8J",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "o4-mini-2025-04-16_medium",
      "score": 0.1896551724137931,
      "bestScore": 0.1896551724137931,
      "standardError": 0.023060480732052885,
      "startedAt": "2025-11-13T15:54:51.979Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XPCVZtFEC6BYKAwhLauMSD",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "o4-mini-2025-04-16_high",
      "score": 0.2482758620689655,
      "bestScore": 0.2482758620689655,
      "standardError": 0.02541251077219606,
      "startedAt": "2025-11-13T15:54:49.087Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Z7hiWJnEkkLmQXviEiTdHt",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5-mini-2025-08-07_medium",
      "score": 0.20344827586206896,
      "bestScore": 0.20344827586206896,
      "standardError": 0.023680184210831092,
      "startedAt": "2025-11-13T15:54:46.227Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "VPHGtT9EdPS6YxbzDAt98q",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "grok-4-0709",
      "score": 0.19655172413793104,
      "bestScore": 0.19655172413793104,
      "standardError": 0.023375906908472147,
      "startedAt": "2025-11-13T15:54:39.607Z",
      "releaseDate": "2025-07-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "oN54TKXxcn9gofYBASTVCY",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5-2025-08-07_medium",
      "score": 0.27241379310344827,
      "bestScore": 0.27241379310344827,
      "standardError": 0.02618833296520288,
      "startedAt": "2025-11-13T15:54:39.070Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hrJBSLQUe5hrHECkuySSJr",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5-nano-2025-08-07_high",
      "score": 0.08275862068965517,
      "bestScore": 0.08275862068965517,
      "standardError": 0.01620688385836769,
      "startedAt": "2025-10-30T16:20:13.144Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "2eMHpXXR9h7P2oNPDQkTBY",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-haiku-4-5-20251001_32K",
      "score": 0.059027777777777776,
      "bestScore": 0.059027777777777776,
      "standardError": 0.013911554772795832,
      "startedAt": "2025-10-22T12:29:30.662Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "fdNzD5mp8r2bHxy2PXx6Gh",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-haiku-4-5-20251001",
      "score": 0.041379310344827586,
      "bestScore": 0.041379310344827586,
      "standardError": 0.011715642254112493,
      "startedAt": "2025-10-16T09:34:31.339Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8sTJmYZ7MSk8obK9vjUjHm",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-5-nano-2025-08-07_medium",
      "score": 0.07241379310344828,
      "bestScore": 0.07241379310344828,
      "standardError": 0.015245401561146893,
      "startedAt": "2025-08-07T18:43:21.321Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dHQzrNGjuR5kiHKtNxG62X",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-20250514_27K",
      "score": 0.041379310344827586,
      "bestScore": 0.041379310344827586,
      "standardError": 0.011715642254112493,
      "startedAt": "2025-08-05T23:13:11.048Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZKroUnGN6qoNssoSQbtzzz",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-1-20250805_27K",
      "score": 0.07241379310344828,
      "bestScore": 0.07241379310344828,
      "standardError": 0.015245401561146893,
      "startedAt": "2025-08-05T19:56:41.845Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ezJ5C38svWJkQMjxTcrgg7",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-1-20250805",
      "score": 0.05862068965517241,
      "bestScore": 0.05862068965517241,
      "standardError": 0.013818435156390344,
      "startedAt": "2025-08-05T19:56:40.259Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "TQQnqmugDk3GX5hkr8HMWz",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-opus-4-20250514",
      "score": 0.04482758620689655,
      "bestScore": 0.04482758620689655,
      "standardError": 0.012172075609474642,
      "startedAt": "2025-07-04T10:42:31.483Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GKD2ELzaXHx3giARevEetr",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-sonnet-4-20250514",
      "score": 0.041379310344827586,
      "bestScore": 0.041379310344827586,
      "standardError": 0.011715642254112493,
      "startedAt": "2025-07-04T10:41:31.264Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "UcfxxJmo4ftRzPEnXrRdSw",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gemini-2.5-pro-preview-06-05",
      "score": 0.10344827586206896,
      "bestScore": 0.10344827586206896,
      "standardError": 0.01791432224407271,
      "startedAt": "2025-07-03T13:57:21.637Z",
      "releaseDate": "2025-06-05",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hNLQNC6TASXNNzat8kVS77",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "qwen-plus-2025-04-28",
      "score": 0.017241379310344827,
      "bestScore": 0.017241379310344827,
      "standardError": 0.007657032895812118,
      "startedAt": "2025-05-14T13:10:51.812Z",
      "releaseDate": "2025-04-28",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "bVFEkuHiUSxjoh8SQm74jo",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "mistral-medium-2505",
      "score": 0.0034602076124567475,
      "bestScore": 0.0034602076124567475,
      "standardError": 0.003460207612456747,
      "startedAt": "2025-05-07T17:33:33.886Z",
      "releaseDate": "2025-05-07",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "PZRX6ME25bn26B6BTRWx6W",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-4.1-nano-2025-04-14",
      "score": 0.010344827586206896,
      "bestScore": 0.010344827586206896,
      "standardError": 0.0059518867144507945,
      "startedAt": "2025-04-14T22:51:44.342Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "chtQvebcoaE33KkhAjbCnw",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-4.1-mini-2025-04-14",
      "score": 0.04482758620689655,
      "bestScore": 0.04482758620689655,
      "standardError": 0.01217207560947464,
      "startedAt": "2025-04-14T22:49:44.216Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ephMMUGDANc9iVTNNuuoZx",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-4.1-2025-04-14",
      "score": 0.05517241379310345,
      "bestScore": 0.05517241379310345,
      "standardError": 0.013430381628597847,
      "startedAt": "2025-04-14T22:40:01.283Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "NPeiYpjbscreYnFkHPvcNB",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "grok-3-mini-beta_low",
      "score": 0.027586206896551724,
      "bestScore": 0.027586206896551724,
      "standardError": 0.009634354634513523,
      "startedAt": "2025-04-10T20:30:21.697Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9FTMS2t6Ksm43CduHeqtJr",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "grok-3-mini-beta_high",
      "score": 0.05862068965517241,
      "bestScore": 0.05862068965517241,
      "standardError": 0.013818435156390344,
      "startedAt": "2025-04-10T18:02:17.992Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Z83F6bJ7Ne5rXaXwgR2xWT",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "grok-3-beta",
      "score": 0.03793103448275862,
      "bestScore": 0.03793103448275862,
      "standardError": 0.011237029601999628,
      "startedAt": "2025-04-10T17:57:03.446Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "DzjFQArKFTZHioQVwKYVtX",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "Llama-4-Maverick-17B-128E-Instruct-FP8",
      "score": 0.006896551724137931,
      "bestScore": 0.006896551724137931,
      "standardError": 0.004868154158215009,
      "startedAt": "2025-04-08T10:17:16.175Z",
      "releaseDate": "2025-04-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "QjvBS34fsngfBBMZUirSyh",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "Llama-4-Scout-17B-16E-Instruct",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-04-08T10:14:56.486Z",
      "releaseDate": "2025-04-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bhjLDA5A7gddsX5cAYKma8",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "qwen-max-2025-01-25",
      "score": 0.010344827586206896,
      "bestScore": 0.010344827586206896,
      "standardError": 0.005951886714450794,
      "startedAt": "2025-04-02T16:55:37.430Z",
      "releaseDate": "2025-01-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "F77vDPynTKJbz7m2bt7Py8",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-3-7-sonnet-20250219_64K",
      "score": 0.03103448275862069,
      "bestScore": 0.03103448275862069,
      "standardError": 0.01020064175308735,
      "startedAt": "2025-03-13T16:49:59.754Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "DoU54Xy93J3MxztC4uxwYy",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-3-7-sonnet-20250219_32K",
      "score": 0.034482758620689655,
      "bestScore": 0.034482758620689655,
      "standardError": 0.010733271038801586,
      "startedAt": "2025-03-13T09:34:11.446Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "aZqVpPa4SWHapN9QHEoEm3",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gemini-2.0-flash-001",
      "score": 0.017241379310344827,
      "bestScore": 0.017241379310344827,
      "standardError": 0.007657032895812118,
      "startedAt": "2025-03-09T17:18:49.858Z",
      "releaseDate": "2025-02-05",
      "organization": "Google DeepMind,Google",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5RpSr8gJZfcZn586aV2hDW",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "DeepSeek-V3",
      "score": 0.017241379310344827,
      "bestScore": 0.017241379310344827,
      "standardError": 0.007657032895812118,
      "startedAt": "2025-03-07T23:36:27.507Z",
      "releaseDate": "2024-12-26",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "aXcfCytWFYNk7E5AZGMzLb",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "o1-2024-12-17_high",
      "score": 0.09310344827586207,
      "bestScore": 0.09310344827586207,
      "standardError": 0.017092785280147835,
      "startedAt": "2025-03-07T19:46:02.827Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bynpttv6jLvbXTY7ktnBpk",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-4o-2024-08-06",
      "score": 0.0034482758620689655,
      "bestScore": 0.0034482758620689655,
      "standardError": 0.003448275862068965,
      "startedAt": "2025-03-07T18:56:32.370Z",
      "releaseDate": "2024-08-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "SmT4ukjkxsXZvFExMMVFgW",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-3-5-sonnet-20240620",
      "score": 0.010344827586206896,
      "bestScore": 0.010344827586206896,
      "standardError": 0.0059518867144507945,
      "startedAt": "2025-03-07T18:54:11.499Z",
      "releaseDate": "2024-06-20",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "h7f9jZVWzZNDHNxP5eewNH",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gemini-1.5-flash-002",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-03-07T18:40:43.945Z",
      "releaseDate": "2024-09-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8FtfWCm8DdAqF9rnMVwv8h",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-3-5-haiku-20241022",
      "score": 0.0034482758620689655,
      "bestScore": 0.0034482758620689655,
      "standardError": 0.003448275862068965,
      "startedAt": "2025-03-07T18:11:31.647Z",
      "releaseDate": "2024-10-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "JQcqpW9xPfJRKJkHne6DpK",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-3-5-sonnet-20241022",
      "score": 0.020689655172413793,
      "bestScore": 0.020689655172413793,
      "standardError": 0.008373130807525473,
      "startedAt": "2025-03-06T23:54:46.118Z",
      "releaseDate": "2024-10-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "MVnDrt94wyokzx3a7ZJkph",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "o1-mini-2024-09-12_medium",
      "score": 0.017241379310344827,
      "bestScore": 0.017241379310344827,
      "standardError": 0.007657032895812118,
      "startedAt": "2025-03-06T22:32:06.577Z",
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hC2FY4WuKm6BLgPhbWeZDY",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "o1-mini-2024-09-12_high",
      "score": 0.013793103448275862,
      "bestScore": 0.013793103448275862,
      "standardError": 0.006860663093423027,
      "startedAt": "2025-03-06T22:30:16.233Z",
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "2ekT2UzaebvmGprKvmeeEa",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-3-7-sonnet-20250219_16K",
      "score": 0.041379310344827586,
      "bestScore": 0.041379310344827586,
      "standardError": 0.011715642254112493,
      "startedAt": "2025-03-06T19:16:31.497Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PU7LXbdNB8Unfvu4qCeJZF",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "grok-2-1212",
      "score": 0.006896551724137931,
      "bestScore": 0.006896551724137931,
      "standardError": 0.004868154158215009,
      "startedAt": "2025-03-06T18:06:22.963Z",
      "releaseDate": "2024-12-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "kinyvFjqSqUiY5FyYv7tsA",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "o3-mini-2025-01-31_medium",
      "score": 0.08081896551724138,
      "bestScore": 0.08081896551724138,
      "standardError": 0.012761417139125066,
      "startedAt": "2025-03-06T18:06:08.042Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "U8CjtcxudQeP2cnMNixfJH",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "claude-3-7-sonnet-20250219",
      "score": 0.03103448275862069,
      "bestScore": 0.03103448275862069,
      "standardError": 0.010200641753087351,
      "startedAt": "2025-03-06T18:05:58.538Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "m7X2ppkiEkweCtFppiRW8S",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "mistral-large-2411",
      "score": 0.0034482758620689655,
      "bestScore": 0.0034482758620689655,
      "standardError": 0.003448275862068965,
      "startedAt": "2025-03-06T18:05:57.178Z",
      "releaseDate": "2024-11-18",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "cEvTUVyxAqq5UDgsGQbvLM",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gpt-4o-2024-11-20",
      "score": 0.0034482758620689655,
      "bestScore": 0.0034482758620689655,
      "standardError": 0.003448275862068965,
      "startedAt": "2025-03-06T18:03:58.731Z",
      "releaseDate": "2024-11-20",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Z7HU95T8eema56EpLtECvG",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-2025-02-28-Private",
      "model": "gemini-2.5-deep-think-2025-08-01-webapp",
      "score": 0.29,
      "bestScore": 0.29,
      "standardError": 0.027,
      "startedAt": null,
      "releaseDate": "2025-08-01",
      "organization": "Google,Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "manual_gemini_2.5_deep_think_run_frontiermath_tiers_1_to_3",
      "sourceFile": "frontiermath.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-8_max",
      "score": 0.3125,
      "bestScore": 0.3125,
      "standardError": 0.06761023307918171,
      "startedAt": "2026-06-08T01:03:04.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5x5EAvuecqo3SXnZWpvrJA",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "qwen3.6-max-preview",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.02914766351555603,
      "startedAt": "2026-05-28T05:51:34.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "6G3krrJBj2Dsuunaow97uR",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gemini-3.5-flash_high",
      "score": 0.14583333333333334,
      "bestScore": 0.14583333333333334,
      "standardError": 0.05148145136372291,
      "startedAt": "2026-05-25T13:24:56.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "icxeMK8LHGTpbEfCNdbqA4",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "qwen3.5-plus",
      "score": 0.020833333333333332,
      "bestScore": 0.020833333333333332,
      "standardError": 0.020833333333333332,
      "startedAt": "2026-05-15T05:49:39.000Z",
      "releaseDate": "2026-02-16",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "ErcX9xv4Pwcgg63Dkrn2zM",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "qwen3.6-plus",
      "score": 0.08333333333333333,
      "bestScore": 0.08333333333333333,
      "standardError": 0.04031495390390805,
      "startedAt": "2026-05-12T19:59:41.000Z",
      "releaseDate": "2026-03-31",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "NuKcKB4qgQwXDPaDEnpqjo",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "qwen3.6-flash",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2026-05-12T16:05:39.000Z",
      "releaseDate": "2026-04-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "4tmJcpwFZKJrmHacCFtPYV",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "qwen3.5-flash",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2026-05-12T13:06:53.000Z",
      "releaseDate": "2026-02-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "cpbqvSFES6AdJqou8FP8ZD",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "glm-5.1",
      "score": 0.125,
      "bestScore": 0.125,
      "standardError": 0.04824031156174749,
      "startedAt": "2026-05-12T05:39:46.000Z",
      "releaseDate": "2026-04-07",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "8vKCYPopFbKzWZ47oLDDSo",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gdm-ai-co-mathematician",
      "score": 0.479,
      "bestScore": 0.479,
      "standardError": 0.072,
      "startedAt": "2026-05-08T13:03:56.784Z",
      "releaseDate": "2026-05-08",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gdm-ai-co-mathematician-manual-run",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "kimi-k2.6",
      "score": 0.1458,
      "bestScore": 0.1458,
      "standardError": 0.0515,
      "startedAt": "2026-05-08T10:29:10.963Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "kimi-k2.6-frontiermth-tier4-frankenstein",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.5-pro-pre-release_xhigh",
      "score": 0.396,
      "bestScore": 0.396,
      "standardError": 0.071,
      "startedAt": "2026-04-23T20:52:04.600Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gpt-5-5-pro-pre-release-xhigh-t4",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.5-pro-pre-release_high",
      "score": 0.396,
      "bestScore": 0.396,
      "standardError": 0.071,
      "startedAt": "2026-04-23T20:51:42.498Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gpt-5-5-pro-pre-release-high-t4",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.5-pre-release_xhigh",
      "score": 0.354,
      "bestScore": 0.354,
      "standardError": 0.069,
      "startedAt": "2026-04-23T20:51:35.635Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gpt-5-5-pre-release-xhigh-t4",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-7_xhigh",
      "score": 0.22916666666666666,
      "bestScore": 0.22916666666666666,
      "standardError": 0.06130657275434798,
      "startedAt": "2026-04-17T12:51:00.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9oFmdEcfQNyYqA5PMmbJxV",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.4-mini-2026-03-17_high",
      "score": 0.0208,
      "bestScore": 0.0208,
      "standardError": 0.0208,
      "startedAt": "2026-04-15T13:54:39.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hPP9Lrd5W3wVyUq9uYuYm9",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.4-nano-2026-03-17_high",
      "score": 0.0625,
      "bestScore": 0.0625,
      "standardError": 0.0353,
      "startedAt": "2026-04-15T09:12:42.664Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZKPz9sjRBK6goz5vzYwtuV",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "muse-spark",
      "score": 0.146,
      "bestScore": 0.146,
      "standardError": 0.051,
      "startedAt": "2026-04-08T16:30:14.687Z",
      "releaseDate": "2026-04-08",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "muse-spark-pre-release-tier-4",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.4-2026-03-05_xhigh",
      "score": 0.271,
      "bestScore": 0.271,
      "standardError": 0.064,
      "startedAt": "2026-03-06T03:53:12.418Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "pre-release-gpt-5-4-xhigh-fm-t4",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.4-pro-2026-03-05-web-app",
      "score": 0.375,
      "bestScore": 0.375,
      "standardError": 0.07,
      "startedAt": "2026-03-06T03:51:41.153Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "pre-release-gpt-5-4-pro-fm-t4-web-app",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-sonnet-4-6_16K",
      "score": 0.083,
      "bestScore": 0.083,
      "standardError": 0.04,
      "startedAt": "2026-02-21T20:36:46.695Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RTfHBvZX3Bh5EPPj6Cf3ab",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gemini-3.1-pro-preview",
      "score": 0.167,
      "bestScore": 0.167,
      "standardError": 0.054,
      "startedAt": "2026-02-19T18:11:11.812Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZZrMMV7TirCbEsHm9pSH27",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "glm-5",
      "score": 0.021,
      "bestScore": 0.021,
      "standardError": 0.021,
      "startedAt": "2026-02-19T17:25:47.564Z",
      "releaseDate": "2026-02-11",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "RomrkYdVrg67NirntgRrDL",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-6_max",
      "score": 0.229,
      "bestScore": 0.229,
      "standardError": 0.061,
      "startedAt": "2026-02-12T16:30:15.979Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "SfehtojAZdZxu7BLgPLDYA",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-6_64K",
      "score": 0.20833333333333334,
      "bestScore": 0.20833333333333334,
      "standardError": 0.05923816975988693,
      "startedAt": "2026-02-06T14:57:13.045Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mAeSEYBeiSU5vUhrnsP6Uj",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-6_32K",
      "score": 0.20833333333333334,
      "bestScore": 0.20833333333333334,
      "standardError": 0.05923816975988693,
      "startedAt": "2026-02-06T12:46:05.800Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "UVXUd6ZFnQTci66adkFepy",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-6",
      "score": 0.14583333333333334,
      "bestScore": 0.14583333333333334,
      "standardError": 0.05148145136372291,
      "startedAt": "2026-02-05T20:01:07.029Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "nvXhiYKL9RDVDfHWw53zFT",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "fireworks/kimi-k2p5",
      "score": 0.042,
      "bestScore": 0.042,
      "standardError": 0.029,
      "startedAt": "2026-02-02T10:58:21.119Z",
      "releaseDate": "2026-01-27",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "F2Fn4eKsSE8ifoE6GUpxC6",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "zai-org/GLM-4.7",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2026-01-30T03:58:10.181Z",
      "releaseDate": "2025-12-22",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "fgfsjZ6NrexyT8kA6arR3B",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.2-pro-2025-12-11-webapp",
      "score": 0.313,
      "bestScore": 0.313,
      "standardError": 0.067,
      "startedAt": "2025-12-24T06:00:00.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "manual_gpt_5.2_pro_run_frontiermath_tier_4",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gemini-2.5-flash",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.02914766351555603,
      "startedAt": "2025-12-18T18:23:16.725Z",
      "releaseDate": "2025-06-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8qSepnktUjAsynVccET4Mw",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gemini-3-flash-preview",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.02914766351555603,
      "startedAt": "2025-12-17T22:40:29.476Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "natXTgCj8Y7mMbV5623YDd",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "fireworks/deepseek-v3p2",
      "score": 0.021,
      "bestScore": 0.021,
      "standardError": 0.021,
      "startedAt": "2025-12-16T15:14:53.837Z",
      "releaseDate": "2025-12-01",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "manual_deepseek-v3.2_frontiermath_tier_4_private_frankenstein_run",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.2-2025-12-11_xhigh",
      "score": 0.188,
      "bestScore": 0.188,
      "standardError": 0.056,
      "startedAt": "2025-12-14T10:55:00.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XLLXdo2tmmvqRwYxbn3kmo",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.2-2025-12-11_high",
      "score": 0.188,
      "bestScore": 0.188,
      "standardError": 0.056,
      "startedAt": "2025-12-11T20:49:18.565Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "TjaWpQnjPC8SNJH8EDcXSS",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.2-2025-12-11_medium",
      "score": 0.167,
      "bestScore": 0.167,
      "standardError": 0.054,
      "startedAt": "2025-12-11T20:49:17.097Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "92XEociAUsCM7WC9knviuo",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.2-2025-12-11_low",
      "score": 0.0625,
      "bestScore": 0.0625,
      "standardError": 0.03530829267912701,
      "startedAt": "2025-12-11T20:22:58.448Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "BvgEDgU5ahio7wQBUMD3oE",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "Qwen/Qwen3-235B-A22B-Thinking-2507",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-12-11T04:59:19.473Z",
      "releaseDate": "2025-07-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "BLw6H5xovDe5FTysit7dNp",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "zai-org/GLM-4.6",
      "score": 0.02127659574468085,
      "bestScore": 0.02127659574468085,
      "standardError": 0.02127659574468085,
      "startedAt": "2025-12-08T19:26:56.763Z",
      "releaseDate": "2025-09-30",
      "organization": "Z.ai (Zhipu AI),Tsinghua University",
      "country": "China",
      "logViewer": null,
      "sourceId": "PDFNt4rT82hpKcKKsA7DEd",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "moonshotai/Kimi-K2-Thinking",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-12-04T16:53:31.304Z",
      "releaseDate": "2025-11-06",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "cnujSZS7UAAQeDu2U5Xfdw",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-5-20251101",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.02914766351555603,
      "startedAt": "2025-11-25T04:02:24.517Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "eAqv27skC4aGnAguQEhUgg",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-5-20251101_16K",
      "score": 0.020833333333333332,
      "bestScore": 0.020833333333333332,
      "standardError": 0.020833333333333332,
      "startedAt": "2025-11-25T01:11:03.782Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "eteYQLCEzxb65iDj6U8din",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-5-20251101_32K",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.02914766351555603,
      "startedAt": "2025-11-25T01:10:48.990Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "BaZfJQ3QBVD9F5cMp7mRdF",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gemini-3-pro-preview",
      "score": 0.1875,
      "bestScore": 0.1875,
      "standardError": 0.05693291125,
      "startedAt": "2025-11-21T12:00:00.000Z",
      "releaseDate": "2025-11-18",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "manual_gemini_3_fmtier4_frankenstein",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.1-2025-11-13_medium",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.029147663515556026,
      "startedAt": "2025-11-17T21:28:56.133Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "S2W9ETQSRNc4xR3cmUjKbF",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5.1-2025-11-13_high",
      "score": 0.125,
      "bestScore": 0.125,
      "standardError": 0.04824031156174749,
      "startedAt": "2025-11-13T20:36:08.563Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ee2YzVv39Tk6tmA7DNjTG5",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5-mini-2025-08-07_high",
      "score": 0.0625,
      "bestScore": 0.0625,
      "standardError": 0.03530829267912701,
      "startedAt": "2025-10-30T16:22:42.856Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5rh6TyPLbkJmQZs8iY6msJ",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5-nano-2025-08-07_high",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-10-30T16:22:14.713Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "CvK572G5SSd5Yk8ZGikDR6",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.125,
      "bestScore": 0.125,
      "standardError": 0.04824031156174749,
      "startedAt": "2025-10-30T02:14:04.513Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6LwyJZsWEzuD8mhSRUF2xn",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-haiku-4-5-20251001_32K",
      "score": 0.020833333333333332,
      "bestScore": 0.020833333333333332,
      "standardError": 0.020833333333333332,
      "startedAt": "2025-10-22T15:24:54.352Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "YRVwm4B88qK6aUEFBdNQmP",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-sonnet-4-5-20250929_32K",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.029147663515556026,
      "startedAt": "2025-10-22T14:19:03.957Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "d8WWwzb6b57mWqddhBXuDi",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "grok-4-heavy-web-app",
      "score": 0.0208,
      "bestScore": 0.0208,
      "standardError": 0.021,
      "startedAt": "2025-10-09T16:00:00.000Z",
      "releaseDate": "2025-07-10",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "grok-4-heavy-manual-run-frontiermath",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5-pro-2025-10-06_high",
      "score": 0.146,
      "bestScore": 0.146,
      "standardError": 0.051,
      "startedAt": "2025-10-09T13:40:00.000Z",
      "releaseDate": "2025-10-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "VshpC5TEp4F72cutXRRmhs",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-sonnet-4-5-20250929",
      "score": 0.020833333333333332,
      "bestScore": 0.020833333333333332,
      "standardError": 0.020833333333333332,
      "startedAt": "2025-09-29T17:41:54.543Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "2o8XZ8V6W5jKD7TuY4frux",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "grok-4-0709",
      "score": 0.020833333333333332,
      "bestScore": 0.020833333333333332,
      "standardError": 0.020833333333333332,
      "startedAt": "2025-08-11T12:47:20.407Z",
      "releaseDate": "2025-07-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Cqdzvvh9tdrdmDjvu8bntM",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "o4-mini-2025-04-16_medium",
      "score": 0.020833333333333332,
      "bestScore": 0.020833333333333332,
      "standardError": 0.020833333333333332,
      "startedAt": "2025-08-07T22:00:02.124Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "oBjz3oZSvAkw5ArLtZMwWW",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5-2025-08-07_medium",
      "score": 0.0625,
      "bestScore": 0.0625,
      "standardError": 0.03530829267912701,
      "startedAt": "2025-08-07T18:43:40.577Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RrmHCttQmYWNaADyThwFwf",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5-mini-2025-08-07_medium",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.02914766351555603,
      "startedAt": "2025-08-07T18:43:35.509Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "EAqUKKCdvdkYTGegACdean",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-5-nano-2025-08-07_medium",
      "score": 0.020833333333333332,
      "bestScore": 0.020833333333333332,
      "standardError": 0.020833333333333332,
      "startedAt": "2025-08-07T18:43:33.699Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "F6Tw2b9fDKde6N9eBu6ory",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-1-20250805_27K",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.029147663515556026,
      "startedAt": "2025-08-05T22:14:34.193Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "a6VpLAmwU23BmxsqrZTGZg",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gemini-2.5-pro",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.02914766351555603,
      "startedAt": "2025-07-03T15:44:08.541Z",
      "releaseDate": "2025-06-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XtEW42hsVgroBKHK465wqA",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gemini-2.5-pro-preview-06-05",
      "score": 0.020833333333333332,
      "bestScore": 0.020833333333333332,
      "standardError": 0.020833333333333332,
      "startedAt": "2025-07-03T13:56:46.043Z",
      "releaseDate": "2025-06-05",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iwLqUvjgt378uQoKiCdSXG",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gpt-4.1-2025-04-14",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-07-01T17:05:38.433Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "oDrEMnoqz3AapWeyrxZd6C",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-20250514",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-07-01T17:05:23.211Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "MUYxn78tdpBMLETSQg5rgN",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "o3-2025-04-16_high",
      "score": 0.020833333333333332,
      "bestScore": 0.020833333333333332,
      "standardError": 0.020833333333333332,
      "startedAt": "2025-07-01T17:05:19.927Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "a9fSMpBem4uhc6pxuEiQqM",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-3-5-sonnet-20240620",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-07-01T17:05:16.422Z",
      "releaseDate": "2024-06-20",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ELsuPViQBj3UPywy6WyB5H",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "o3-mini-2025-01-31_high",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.029147663515556026,
      "startedAt": "2025-07-01T17:05:14.302Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "YZ6Ts7tSVrvzjSRGxJowdn",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-opus-4-20250514_27K",
      "score": 0.041666666666666664,
      "bestScore": 0.041666666666666664,
      "standardError": 0.029147663515556026,
      "startedAt": "2025-07-01T17:05:12.990Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Ee6YnzHKyAUxmYeL23TL9s",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-3-5-sonnet-20241022",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-07-01T17:05:10.267Z",
      "releaseDate": "2024-10-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8mhf6cwsHtrRkAwxZSBgfV",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "o4-mini-2025-04-16_high",
      "score": 0.0625,
      "bestScore": 0.0625,
      "standardError": 0.03530829267912701,
      "startedAt": "2025-07-01T17:05:06.108Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "WgEkfyLvjHYXfPje5Z2CPX",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "grok-3-beta",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-07-01T17:05:01.787Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZyHLq7azxVkZJZRLTj3kMB",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "claude-sonnet-4-20250514",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-07-01T17:04:25.401Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XMYUtk8tkqDFgohY8iYfmy",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-2025-07-01-Private",
      "model": "gemini-2.5-deep-think-2025-08-01-webapp",
      "score": 0.104,
      "bestScore": 0.104,
      "standardError": 0.044,
      "startedAt": null,
      "releaseDate": "2025-08-01",
      "organization": "Google,Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "manual_gemini_2.5_deep_think_run_frontiermath_tier_4",
      "sourceFile": "frontiermath_tier_4.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "Muse-Spark-Open-816B-A42B_xhigh",
      "score": 0.6982456140350877,
      "bestScore": 0.6982456140350877,
      "standardError": 0.027237781633151586,
      "startedAt": "2026-10-15T18:52:04.000Z",
      "releaseDate": "2026-10-15",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "e3e3h9r7Fouu4wsAeVzTnV",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-4-argon_high",
      "score": 0.8596491228070176,
      "bestScore": 0.8596491228070176,
      "standardError": 0.020611471958164294,
      "startedAt": "2026-10-13T17:14:57.000Z",
      "releaseDate": "2026-09-30",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "N8daAyKM6Sah39YBcPRi4W",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "mistral-large-4-0_high",
      "score": 0.5859649122807018,
      "bestScore": 0.5859649122807018,
      "standardError": 0.02922773823944577,
      "startedAt": "2026-10-12T21:38:35.000Z",
      "releaseDate": "2026-10-06",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "CAstp8hQG6Aehvtw85bMDX",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "muse-wonder-1.0_max",
      "score": 0.8210526315789474,
      "bestScore": 0.8210526315789474,
      "standardError": 0.02274515950332741,
      "startedAt": "2026-10-12T19:00:20.000Z",
      "releaseDate": "2026-10-12",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hwVgELmPofPY3Vqm7BRLxs",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "deepseek-v4.1-flash_max",
      "score": 0.6736842105263158,
      "bestScore": 0.6736842105263158,
      "standardError": 0.027821981570324446,
      "startedAt": "2026-10-08T19:02:32.000Z",
      "releaseDate": "2026-09-09",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "b6dXtaGSwvfRYADKYnj9Ci",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-haiku-5-5_max",
      "score": 0.7508771929824561,
      "bestScore": 0.7508771929824561,
      "standardError": 0.02566445410232946,
      "startedAt": "2026-10-07T20:42:31.000Z",
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "2QAt4ptgGkLtcbVL3hdAvL",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-6.1-sol_max",
      "score": 0.9368421052631579,
      "bestScore": 0.9368421052631579,
      "standardError": 0.014434038295343787,
      "startedAt": "2026-09-29T22:00:34.000Z",
      "releaseDate": "2026-09-29",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "aBoQLaqnXsLPrsbN8wYgUp",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-sonnet-5-5_max",
      "score": 0.887719298245614,
      "bestScore": 0.887719298245614,
      "standardError": 0.018734030360524248,
      "startedAt": "2026-09-29T15:15:14.000Z",
      "releaseDate": "2026-09-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4tZWwV9tqEgsCaKeSuSoG4",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-6-luna_max",
      "score": 0.7894736842105263,
      "bestScore": 0.7894736842105263,
      "standardError": 0.024191503146153773,
      "startedAt": "2026-09-22T22:14:05.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iBQxysYR7zPxB6hxvJyTLc",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "grok-4.7_xhigh",
      "score": 0.5298245614035088,
      "bestScore": 0.5298245614035088,
      "standardError": 0.029616712041900365,
      "startedAt": "2026-09-22T21:49:22.000Z",
      "releaseDate": "2026-09-21",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "F4unBbW5M4VJogzCq4vGAb",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-6-sol_max",
      "score": 0.8982456140350877,
      "bestScore": 0.8982456140350877,
      "standardError": 0.01793969072128637,
      "startedAt": "2026-09-22T19:59:24.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RHjHEPHEsWZqFuMjBTeC6r",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-opus-5-5_max",
      "score": 0.9122807017543859,
      "bestScore": 0.9122807017543859,
      "standardError": 0.016786209858442095,
      "startedAt": "2026-09-22T18:27:39.000Z",
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LPxhvBNEn7S3xkRZUgcvc9",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "muse-spark-1.3_max",
      "score": 0.7403508771929824,
      "bestScore": 0.7403508771929824,
      "standardError": 0.026016750822385526,
      "startedAt": "2026-09-18T16:34:54.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6735DShGR4Zu5KZ2U6n4pC",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "muse-spark-1.3_xhigh",
      "score": 0.743859649122807,
      "bestScore": 0.743859649122807,
      "standardError": 0.025901524747773286,
      "startedAt": "2026-09-16T22:00:25.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "JeYksd3f2RkegmxCcZyBMQ",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.8-max-0902_xhigh",
      "score": 0.656140350877193,
      "bestScore": 0.656140350877193,
      "standardError": 0.02818576398907258,
      "startedAt": "2026-09-02T18:04:58.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "FpQf5xBwJPb4ZEo6SgdKWq",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-3.8-flash_high",
      "score": 0.6842105263157895,
      "bestScore": 0.6842105263157895,
      "standardError": 0.027582557383453006,
      "startedAt": "2026-09-02T17:15:36.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Utgs6DuaQDDdi98ZWH22qL",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-fable-5-1_max",
      "score": 0.9017543859649123,
      "bestScore": 0.9017543859649123,
      "standardError": 0.017662067789031757,
      "startedAt": "2026-09-01T18:38:24.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "g7dXwZH2tWvqLg3X5idGs2",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-3.1-flash-lite_low",
      "score": 0.22456140350877193,
      "bestScore": 0.22456140350877193,
      "standardError": 0.024761793289496806,
      "startedAt": "2026-08-30T22:35:03.000Z",
      "releaseDate": "2026-03-03",
      "organization": "Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVcuyQvF8LWuYhQCx2FGDHh.eval",
      "sourceId": "VcuyQvF8LWuYhQCx2FGDHh",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-3.1-flash-lite_high",
      "score": 0.2771929824561403,
      "bestScore": 0.2771929824561403,
      "standardError": 0.026560920735701255,
      "startedAt": "2026-08-30T22:34:53.000Z",
      "releaseDate": "2026-03-03",
      "organization": "Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMVZf59jt9uEoWG9w5W9apB.eval",
      "sourceId": "MVZf59jt9uEoWG9w5W9apB",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.6-35b-a3b_none",
      "score": 0.20350877192982456,
      "bestScore": 0.20350877192982456,
      "standardError": 0.02389037106974561,
      "startedAt": "2026-08-30T22:34:37.000Z",
      "releaseDate": "2026-04-14",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYRvn36dmMyDbS5e4XkF4Xi.eval",
      "sourceId": "YRvn36dmMyDbS5e4XkF4Xi",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.6-plus",
      "score": 0.3824561403508772,
      "bestScore": 0.3824561403508772,
      "standardError": 0.028838027580159496,
      "startedAt": "2026-08-30T22:34:20.000Z",
      "releaseDate": "2026-03-31",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjbyVWnEvvFVNNc8AEF6WcQ.eval",
      "sourceId": "jbyVWnEvvFVNNc8AEF6WcQ",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.5-397b-a17b_none",
      "score": 0.312280701754386,
      "bestScore": 0.312280701754386,
      "standardError": 0.027499133473298815,
      "startedAt": "2026-08-30T22:34:18.000Z",
      "releaseDate": "2026-02-13",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FA4yUA9PfzRr92JAqDDo9ro.eval",
      "sourceId": "A4yUA9PfzRr92JAqDDo9ro",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3-max-2025-09-23",
      "score": 0.18947368421052632,
      "bestScore": 0.18947368421052632,
      "standardError": 0.023254062737737245,
      "startedAt": "2026-08-30T22:34:08.000Z",
      "releaseDate": "2025-09-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FNrTH3EYHjp4hPdxgpCPGTX.eval",
      "sourceId": "NrTH3EYHjp4hPdxgpCPGTX",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-6-astra_max",
      "score": 0.9368421052631579,
      "bestScore": 0.9368421052631579,
      "standardError": 0.014434038295343787,
      "startedAt": "2026-08-30T14:57:43.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "csK2fe5T8p2R7WgoJUu6bZ",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "glm-5.2_none",
      "score": 0.4245614035087719,
      "bestScore": 0.4245614035087719,
      "standardError": 0.029329899789943815,
      "startedAt": "2026-08-29T11:07:59.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FECdG3m8n5ABfWVZ55rPFmT.eval",
      "sourceId": "ECdG3m8n5ABfWVZ55rPFmT",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.7-flash_none",
      "score": 0.19298245614035087,
      "bestScore": 0.19298245614035087,
      "standardError": 0.02341753795065531,
      "startedAt": "2026-08-29T11:07:51.000Z",
      "releaseDate": "2026-07-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdoNrgGQY5SytQV22WRqw5L.eval",
      "sourceId": "doNrgGQY5SytQV22WRqw5L",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.7-flash",
      "score": 0.19298245614035087,
      "bestScore": 0.19298245614035087,
      "standardError": 0.023417537950655314,
      "startedAt": "2026-08-29T11:07:49.000Z",
      "releaseDate": "2026-07-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6pTYmQCw3EMbErvAqpivH5.eval",
      "sourceId": "6pTYmQCw3EMbErvAqpivH5",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.6-luna_none",
      "score": 0.39649122807017545,
      "bestScore": 0.39649122807017545,
      "standardError": 0.029026817016410068,
      "startedAt": "2026-08-29T11:07:39.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYwHKNFQjbat97wVLqachmi.eval",
      "sourceId": "YwHKNFQjbat97wVLqachmi",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.7-plus_none",
      "score": 0.34385964912280703,
      "bestScore": 0.34385964912280703,
      "standardError": 0.028185763989072583,
      "startedAt": "2026-08-29T11:07:26.000Z",
      "releaseDate": "2026-06-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTt8tBXHoCEYqGLePWXkM8u.eval",
      "sourceId": "Tt8tBXHoCEYqGLePWXkM8u",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.6-luna_low",
      "score": 0.41403508771929826,
      "bestScore": 0.41403508771929826,
      "standardError": 0.029227738239445765,
      "startedAt": "2026-08-29T11:07:20.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZryXhNUpPedPvGhbqdNBEw.eval",
      "sourceId": "ZryXhNUpPedPvGhbqdNBEw",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "glm-5.2_low",
      "score": 0.5473684210526316,
      "bestScore": 0.5473684210526316,
      "standardError": 0.0295360982699221,
      "startedAt": "2026-08-29T11:07:19.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FUnKHyc2idtTmoTUGxGeNJV.eval",
      "sourceId": "UnKHyc2idtTmoTUGxGeNJV",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.6-flash",
      "score": 0.17192982456140352,
      "bestScore": 0.17192982456140352,
      "standardError": 0.022389789548206734,
      "startedAt": "2026-08-29T11:07:18.000Z",
      "releaseDate": "2026-04-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FnDsNFM5ioGzyWcpeUKVotg.eval",
      "sourceId": "nDsNFM5ioGzyWcpeUKVotg",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.6-flash_none",
      "score": 0.22456140350877193,
      "bestScore": 0.22456140350877193,
      "standardError": 0.024761793289496806,
      "startedAt": "2026-08-29T11:06:46.000Z",
      "releaseDate": "2026-04-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRoscWx4NUGVqcHdko6QhKd.eval",
      "sourceId": "RoscWx4NUGVqcHdko6QhKd",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.6-plus_none",
      "score": 0.32280701754385965,
      "bestScore": 0.32280701754385965,
      "standardError": 0.02774396565327428,
      "startedAt": "2026-08-29T11:06:40.000Z",
      "releaseDate": "2026-03-31",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZ4xfwTke5pR6RrY952bttA.eval",
      "sourceId": "Z4xfwTke5pR6RrY952bttA",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.6-35b-a3b",
      "score": 0.17543859649122806,
      "bestScore": 0.17543859649122806,
      "standardError": 0.022569134425265196,
      "startedAt": "2026-08-29T11:06:39.000Z",
      "releaseDate": "2026-04-14",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FggG3n6WVPZn7Gu2RNyfQbK.eval",
      "sourceId": "ggG3n6WVPZn7Gu2RNyfQbK",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.6-27b_none",
      "score": 0.34035087719298246,
      "bestScore": 0.34035087719298246,
      "standardError": 0.028116467881852295,
      "startedAt": "2026-08-29T11:06:28.000Z",
      "releaseDate": "2026-04-22",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWjXdTPMnJUAT27azUKo3Ka.eval",
      "sourceId": "WjXdTPMnJUAT27azUKo3Ka",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "glm-5.1",
      "score": 0.3684210526315789,
      "bestScore": 0.3684210526315789,
      "standardError": 0.028623772537150215,
      "startedAt": "2026-08-29T11:06:22.000Z",
      "releaseDate": "2026-04-07",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMhwme9odWJMzMjoYkW8r6B.eval",
      "sourceId": "Mhwme9odWJMzMjoYkW8r6B",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-3.1-flash-lite_minimal",
      "score": 0.21403508771929824,
      "bestScore": 0.21403508771929824,
      "standardError": 0.02433800055326239,
      "startedAt": "2026-08-28T12:55:30.000Z",
      "releaseDate": "2026-03-03",
      "organization": "Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F4gZy2dAAT2EPZDGz5puD3H.eval",
      "sourceId": "4gZy2dAAT2EPZDGz5puD3H",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.6-27b",
      "score": 0.3508771929824561,
      "bestScore": 0.3508771929824561,
      "standardError": 0.028319255656547093,
      "startedAt": "2026-08-28T12:30:22.000Z",
      "releaseDate": "2026-04-22",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVonoEhAsGSJ2FHU33qnhui.eval",
      "sourceId": "VonoEhAsGSJ2FHU33qnhui",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "glm-5.1_none",
      "score": 0.24912280701754386,
      "bestScore": 0.24912280701754386,
      "standardError": 0.02566445410232946,
      "startedAt": "2026-08-28T12:30:03.000Z",
      "releaseDate": "2026-04-07",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F4hmmfwG9tgZJvKhL78nKzL.eval",
      "sourceId": "4hmmfwG9tgZJvKhL78nKzL",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.4-nano-2026-03-17_none",
      "score": 0.0456140350877193,
      "bestScore": 0.0456140350877193,
      "standardError": 0.012380896644748405,
      "startedAt": "2026-08-28T12:29:12.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKDQZWiPXB2VXc4bj4GJQW4.eval",
      "sourceId": "KDQZWiPXB2VXc4bj4GJQW4",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.4-mini-2026-03-17_none",
      "score": 0.17192982456140352,
      "bestScore": 0.17192982456140352,
      "standardError": 0.022389789548206734,
      "startedAt": "2026-08-28T12:29:04.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgDwrSBE5GvFXymdBYMnMyV.eval",
      "sourceId": "gDwrSBE5GvFXymdBYMnMyV",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.4-nano-2026-03-17_low",
      "score": 0.20350877192982456,
      "bestScore": 0.20350877192982456,
      "standardError": 0.02389037106974561,
      "startedAt": "2026-08-28T12:28:58.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FBNSjwbcpNSaDXHF7gC9WoN.eval",
      "sourceId": "BNSjwbcpNSaDXHF7gC9WoN",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.4-mini-2026-03-17_low",
      "score": 0.24561403508771928,
      "bestScore": 0.24561403508771928,
      "standardError": 0.0255425481026507,
      "startedAt": "2026-08-28T12:28:53.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F8nsNpE6ETikLf4tSrNHmGs.eval",
      "sourceId": "8nsNpE6ETikLf4tSrNHmGs",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.5-flash_none",
      "score": 0.1824561403508772,
      "bestScore": 0.1824561403508772,
      "standardError": 0.022917941277334045,
      "startedAt": "2026-08-28T12:28:39.000Z",
      "releaseDate": "2026-02-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FafxoicJc4C6caVDwUpsiME.eval",
      "sourceId": "afxoicJc4C6caVDwUpsiME",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.5-flash",
      "score": 0.09473684210526316,
      "bestScore": 0.09473684210526316,
      "standardError": 0.017377516319276178,
      "startedAt": "2026-08-28T12:28:30.000Z",
      "releaseDate": "2026-02-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6XZAenohp9pCgQ3gVeUAi8.eval",
      "sourceId": "6XZAenohp9pCgQ3gVeUAi8",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.5-397b-a17b",
      "score": 0.29473684210526313,
      "bestScore": 0.29473684210526313,
      "standardError": 0.02705413430767295,
      "startedAt": "2026-08-28T12:28:11.000Z",
      "releaseDate": "2026-02-13",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FEzMaDdXs4YQTMLgppBGxAb.eval",
      "sourceId": "EzMaDdXs4YQTMLgppBGxAb",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o1-2024-12-17_high",
      "score": 0.14736842105263157,
      "bestScore": 0.14736842105263157,
      "standardError": 0.021034091168852485,
      "startedAt": "2026-08-28T07:09:23.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FNMJhDtxtyQJCHQ4EApr9LM.eval",
      "sourceId": "NMJhDtxtyQJCHQ4EApr9LM",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o1-2024-12-17_low",
      "score": 0.08421052631578947,
      "bestScore": 0.08421052631578947,
      "standardError": 0.016478658153935063,
      "startedAt": "2026-08-27T21:00:29.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQnMPQZzDRHcTKZRaWYunaa.eval",
      "sourceId": "QnMPQZzDRHcTKZRaWYunaa",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o1-2024-12-17_medium",
      "score": 0.10175438596491228,
      "bestScore": 0.10175438596491228,
      "standardError": 0.01793969072128637,
      "startedAt": "2026-08-27T21:00:05.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F4RnNJhaFMGdLSFjoHHMR8k.eval",
      "sourceId": "4RnNJhaFMGdLSFjoHHMR8k",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o3-2025-04-16_high",
      "score": 0.3333333333333333,
      "bestScore": 0.3333333333333333,
      "standardError": 0.02797271194322297,
      "startedAt": "2026-08-27T20:00:08.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FN4HerYebFnJDEU4PEvNNGM.eval",
      "sourceId": "N4HerYebFnJDEU4PEvNNGM",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o3-2025-04-16_medium",
      "score": 0.2982456140350877,
      "bestScore": 0.2982456140350877,
      "standardError": 0.027146911721220194,
      "startedAt": "2026-08-27T19:59:58.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQV5JgKBrZtfVmNj9fRWJKA.eval",
      "sourceId": "QV5JgKBrZtfVmNj9fRWJKA",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5-2025-08-07_low",
      "score": 0.3719298245614035,
      "bestScore": 0.3719298245614035,
      "standardError": 0.028679753751074614,
      "startedAt": "2026-08-27T19:59:52.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJ2DWDpGLBhihZx82M3hhug.eval",
      "sourceId": "J2DWDpGLBhihZx82M3hhug",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-4.1-mini-2025-04-14",
      "score": 0.06666666666666667,
      "bestScore": 0.06666666666666667,
      "standardError": 0.014801767859562859,
      "startedAt": "2026-08-27T19:59:44.000Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6EQrp2MY4QMYLfTxpioYTe.eval",
      "sourceId": "6EQrp2MY4QMYLfTxpioYTe",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-3.5-turbo-0125",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2026-08-27T19:59:25.000Z",
      "releaseDate": "2024-01-25",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFNaBAvmeTiRRFDW63KemZ7.eval",
      "sourceId": "FNaBAvmeTiRRFDW63KemZ7",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-4o-mini-2024-07-18",
      "score": 0.007017543859649123,
      "bestScore": 0.007017543859649123,
      "standardError": 0.004953408961963604,
      "startedAt": "2026-08-27T19:51:52.000Z",
      "releaseDate": "2024-07-18",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6XhtKr59amKko65T7RUB2a.eval",
      "sourceId": "6XhtKr59amKko65T7RUB2a",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o3-2025-04-16_low",
      "score": 0.19298245614035087,
      "bestScore": 0.19298245614035087,
      "standardError": 0.023417537950655314,
      "startedAt": "2026-08-27T17:55:20.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCX3BTd8Yir7nQaRaNGWhA4.eval",
      "sourceId": "CX3BTd8Yir7nQaRaNGWhA4",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-4.1-2025-04-14",
      "score": 0.05964912280701754,
      "bestScore": 0.05964912280701754,
      "standardError": 0.014053607746337793,
      "startedAt": "2026-08-27T17:54:56.000Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FkCG2nzspryMy7LBwWPqwXy.eval",
      "sourceId": "kCG2nzspryMy7LBwWPqwXy",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-4-turbo-2024-04-09",
      "score": 0.007017543859649123,
      "bestScore": 0.007017543859649123,
      "standardError": 0.004953408961963604,
      "startedAt": "2026-08-27T17:54:49.000Z",
      "releaseDate": "2024-04-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbC2dbN4x2GWWmnnfseKxgQ.eval",
      "sourceId": "bC2dbN4x2GWWmnnfseKxgQ",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-4o-2024-08-06",
      "score": 0.0035087719298245615,
      "bestScore": 0.0035087719298245615,
      "standardError": 0.0035087719298245615,
      "startedAt": "2026-08-27T17:54:29.000Z",
      "releaseDate": "2024-08-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FikS95CXY2foitnUPDKwNJn.eval",
      "sourceId": "ikS95CXY2foitnUPDKwNJn",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o4-mini-2025-04-16_low",
      "score": 0.16140350877192983,
      "bestScore": 0.16140350877192983,
      "standardError": 0.021831010387413353,
      "startedAt": "2026-08-27T17:54:12.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCpEZwR38bY8xb29bVmNHFm.eval",
      "sourceId": "CpEZwR38bY8xb29bVmNHFm",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o4-mini-2025-04-16_medium",
      "score": 0.28771929824561404,
      "bestScore": 0.28771929824561404,
      "standardError": 0.02686277751301251,
      "startedAt": "2026-08-27T17:53:57.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWorgtXwbnWQSAzMfPgbBH9.eval",
      "sourceId": "WorgtXwbnWQSAzMfPgbBH9",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o3-mini-2025-01-31_medium",
      "score": 0.10526315789473684,
      "bestScore": 0.10526315789473684,
      "standardError": 0.01821070199635025,
      "startedAt": "2026-08-27T17:53:55.000Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhRYELqPG6rnY5cBtJhYnt7.eval",
      "sourceId": "hRYELqPG6rnY5cBtJhYnt7",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5-mini-2025-08-07_minimal",
      "score": 0.05964912280701754,
      "bestScore": 0.05964912280701754,
      "standardError": 0.014053607746337795,
      "startedAt": "2026-08-27T17:53:38.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fjrbb2oJv95wiFbnCfN56y9.eval",
      "sourceId": "jrbb2oJv95wiFbnCfN56y9",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5-2025-08-07_minimal",
      "score": 0.1824561403508772,
      "bestScore": 0.1824561403508772,
      "standardError": 0.022917941277334045,
      "startedAt": "2026-08-27T17:53:33.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F53PtW8g64r6zeLY38aK9Q4.eval",
      "sourceId": "53PtW8g64r6zeLY38aK9Q4",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5-mini-2025-08-07_low",
      "score": 0.1824561403508772,
      "bestScore": 0.1824561403508772,
      "standardError": 0.022917941277334045,
      "startedAt": "2026-08-27T17:53:29.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSrVpkvKRweaBHSVwPLAhBQ.eval",
      "sourceId": "SrVpkvKRweaBHSVwPLAhBQ",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5-nano-2025-08-07_minimal",
      "score": 0.017543859649122806,
      "bestScore": 0.017543859649122806,
      "standardError": 0.00779040413661203,
      "startedAt": "2026-08-27T17:53:21.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FnFt6PoshWAVXE6iH9PQgo2.eval",
      "sourceId": "nFt6PoshWAVXE6iH9PQgo2",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o3-mini-2025-01-31_low",
      "score": 0.03859649122807018,
      "bestScore": 0.03859649122807018,
      "standardError": 0.011430562271752924,
      "startedAt": "2026-08-27T17:53:02.000Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FabYpn5zgPpRdBoYdzuinPt.eval",
      "sourceId": "abYpn5zgPpRdBoYdzuinPt",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5-nano-2025-08-07_low",
      "score": 0.05964912280701754,
      "bestScore": 0.05964912280701754,
      "standardError": 0.014053607746337795,
      "startedAt": "2026-08-27T17:52:46.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FkfDcrjxJgjsW8CvG3L8yxz.eval",
      "sourceId": "kfDcrjxJgjsW8CvG3L8yxz",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "glm-5.3-flash_max",
      "score": 0.5578947368421052,
      "bestScore": 0.5578947368421052,
      "standardError": 0.029469977528223976,
      "startedAt": "2026-08-27T13:29:52.000Z",
      "releaseDate": "2026-08-20",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "YC2u9wykbYQg4gdkf76j26",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "glm-5.3_max",
      "score": 0.6877192982456141,
      "bestScore": 0.6877192982456141,
      "standardError": 0.027499133473298815,
      "startedAt": "2026-08-25T14:20:20.000Z",
      "releaseDate": "2026-08-14",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "iXvqUzwxAiA3Rg3dwnutFs",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "deepseek-v4-pro-0813_max",
      "score": 0.6456140350877193,
      "bestScore": 0.6456140350877193,
      "standardError": 0.028383475205435625,
      "startedAt": "2026-08-19T13:35:48.000Z",
      "releaseDate": "2026-08-13",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fb6yWmeFgsWeEiG9Qzs3NSf.eval",
      "sourceId": "b6yWmeFgsWeEiG9Qzs3NSf",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-3.7-flash_high",
      "score": 0.7157894736842105,
      "bestScore": 0.7157894736842105,
      "standardError": 0.026764156649364653,
      "startedAt": "2026-08-14T19:59:58.000Z",
      "releaseDate": "2026-08-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FaxMdgNDC8UYYLnZYLHZpF5.eval",
      "sourceId": "axMdgNDC8UYYLnZYLHZpF5",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "inkling-small_xhigh",
      "score": 0.4631578947368421,
      "bestScore": 0.4631578947368421,
      "standardError": 0.02958888847874604,
      "startedAt": "2026-08-14T19:59:09.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FB6bZbqickq8o2fQcgDmzMo.eval",
      "sourceId": "B6bZbqickq8o2fQcgDmzMo",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "grok-4.6_xhigh",
      "score": 0.6596491228070176,
      "bestScore": 0.6596491228070176,
      "standardError": 0.028116467881852295,
      "startedAt": "2026-08-14T15:46:50.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZb2AKwa942D7dBTTcebAba.eval",
      "sourceId": "Zb2AKwa942D7dBTTcebAba",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "Inkling_xhigh",
      "score": 0.3333333333333333,
      "bestScore": 0.3333333333333333,
      "standardError": 0.02797271194322297,
      "startedAt": "2026-08-06T20:14:32.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6KvQvuMMX5zAuteUc6pFcW.eval",
      "sourceId": "6KvQvuMMX5zAuteUc6pFcW",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.8-max_xhigh",
      "score": 0.7473684210526316,
      "bestScore": 0.7473684210526316,
      "standardError": 0.0257841025556124,
      "startedAt": "2026-08-04T19:32:51.000Z",
      "releaseDate": "2026-08-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FABT89ieSWGnGj4RYMduHiG.eval",
      "sourceId": "ABT89ieSWGnGj4RYMduHiG",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.5-instant",
      "score": 0.2631578947368421,
      "bestScore": 0.2631578947368421,
      "standardError": 0.026129809832490088,
      "startedAt": "2026-08-02T03:23:36.000Z",
      "releaseDate": "2026-05-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgGcHjE6ppy54ahN2pfxsGH.eval",
      "sourceId": "gGcHjE6ppy54ahN2pfxsGH",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "deepseek-v4-flash-0731_max",
      "score": 0.5754385964912281,
      "bestScore": 0.5754385964912281,
      "standardError": 0.029329899789943818,
      "startedAt": "2026-08-02T02:15:37.000Z",
      "releaseDate": "2026-07-31",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmKhVMEicZeGjdvJdTRd7mK.eval",
      "sourceId": "mKhVMEicZeGjdvJdTRd7mK",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-3.5-flash-lite_high",
      "score": 0.2596491228070175,
      "bestScore": 0.2596491228070175,
      "standardError": 0.02601675082238553,
      "startedAt": "2026-08-02T01:20:17.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F8o2J9uQLKDCTiCFwYdkJRK.eval",
      "sourceId": "8o2J9uQLKDCTiCFwYdkJRK",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-3.6-flash_high",
      "score": 0.5894736842105263,
      "bestScore": 0.5894736842105263,
      "standardError": 0.02919063494391404,
      "startedAt": "2026-08-02T00:36:45.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fnb9LnDeeQyvRCqsbsTANGD.eval",
      "sourceId": "nb9LnDeeQyvRCqsbsTANGD",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-opus-5_max",
      "score": 0.856140350877193,
      "bestScore": 0.856140350877193,
      "standardError": 0.020824894575343943,
      "startedAt": "2026-07-24T19:54:20.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fff7GeS7rdperjxSBZ7MHZY.eval",
      "sourceId": "ff7GeS7rdperjxSBZ7MHZY",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "kimi-k3_max",
      "score": 0.721830985915493,
      "bestScore": 0.721830985915493,
      "standardError": 0.026636607935112643,
      "startedAt": "2026-07-17T01:48:42.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "bcntamvGsSH6QgETdoaSph",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "grok-4.20-0309-reasoning",
      "score": 0.44912280701754387,
      "bestScore": 0.44912280701754387,
      "standardError": 0.029515543245521678,
      "startedAt": "2026-07-13T15:51:35.000Z",
      "releaseDate": "2026-02-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFLDt53waLTeshC7PK9uVy2.eval",
      "sourceId": "FLDt53waLTeshC7PK9uVy2",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.6-terra_max",
      "score": 0.8596491228070176,
      "bestScore": 0.8596491228070176,
      "standardError": 0.020611471958164294,
      "startedAt": "2026-07-09T15:58:30.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJ6evjTMgHkrun8txnEvzdN.eval",
      "sourceId": "J6evjTMgHkrun8txnEvzdN",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.6-luna_max",
      "score": 0.8210526315789474,
      "bestScore": 0.8210526315789474,
      "standardError": 0.02274515950332741,
      "startedAt": "2026-07-09T14:18:35.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYLAk5J3Wd8EruVXee8bAQb.eval",
      "sourceId": "YLAk5J3Wd8EruVXee8bAQb",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.6-sol_max",
      "score": 0.8912280701754386,
      "bestScore": 0.8912280701754386,
      "standardError": 0.018475392571140503,
      "startedAt": "2026-07-09T02:45:09.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVBpT2WWYKBZtUxG9Z3svGh.eval",
      "sourceId": "VBpT2WWYKBZtUxG9Z3svGh",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "grok-4.5_high",
      "score": 0.5719298245614035,
      "bestScore": 0.5719298245614035,
      "standardError": 0.029360921879029937,
      "startedAt": "2026-07-09T01:53:42.000Z",
      "releaseDate": "2026-07-08",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "djbbEqvBW9qux3pZwcZN2i",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-sonnet-5_max",
      "score": 0.656140350877193,
      "bestScore": 0.656140350877193,
      "standardError": 0.02818576398907258,
      "startedAt": "2026-06-30T21:54:08.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FV6YuGpcu8MCbJNXKoJMA6b.eval",
      "sourceId": "V6YuGpcu8MCbJNXKoJMA6b",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "glm-5.2_max",
      "score": 0.592057761732852,
      "bestScore": 0.592057761732852,
      "standardError": 0.029581952519606228,
      "startedAt": "2026-06-19T16:13:05.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "TTuCPacMDpNaCSs3Hpbdjp",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "grok-4.3_high",
      "score": 0.4280701754385965,
      "bestScore": 0.4280701754385965,
      "standardError": 0.029360921879029937,
      "startedAt": "2026-06-17T17:10:48.000Z",
      "releaseDate": "2026-04-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbdsZjrG4ayzP8NAJg5q586.eval",
      "sourceId": "bdsZjrG4ayzP8NAJg5q586",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "deepseek-v4-pro_max",
      "score": 0.45263157894736844,
      "bestScore": 0.45263157894736844,
      "standardError": 0.029536098269922092,
      "startedAt": "2026-06-17T02:30:14.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKUar3FaFjzq5SMKrEhtn8e.eval",
      "sourceId": "KUar3FaFjzq5SMKrEhtn8e",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "qwen3.7-max",
      "score": 0.6456140350877193,
      "bestScore": 0.6456140350877193,
      "standardError": 0.02838347520543562,
      "startedAt": "2026-06-13T03:08:22.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRG2biwnmGybSVTVxxMdRWa.eval",
      "sourceId": "RG2biwnmGybSVTVxxMdRWa",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.4-pro-2026-03-05_xhigh",
      "score": 0.8245614035087719,
      "bestScore": 0.8245614035087719,
      "standardError": 0.022569134425265196,
      "startedAt": "2026-06-13T02:38:02.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZFVa8dw7oiir6PhcF8XZMo.eval",
      "sourceId": "ZFVa8dw7oiir6PhcF8XZMo",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.2-pro-2025-12-11_xhigh",
      "score": 0.74,
      "bestScore": 0.74,
      "standardError": 0.026349537704451795,
      "startedAt": "2026-06-13T02:37:55.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dMJenf5xYWu7Pwn2U6w4dW",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "kimi-k2.7-code",
      "score": 0.5403508771929825,
      "bestScore": 0.5403508771929825,
      "standardError": 0.02957276813514758,
      "startedAt": "2026-06-13T01:39:44.000Z",
      "releaseDate": "2026-06-12",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FX3RxRkt2pBUXYJjAQHsUo6.eval",
      "sourceId": "X3RxRkt2pBUXYJjAQHsUo6",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.5-pro_xhigh",
      "score": 0.8771929824561403,
      "bestScore": 0.8771929824561403,
      "standardError": 0.019476010341530077,
      "startedAt": "2026-06-12T21:33:12.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LSKDetgzyVojru45w7RJhf",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5-pro-2025-10-06_high",
      "score": 0.5578947368421052,
      "bestScore": 0.5578947368421052,
      "standardError": 0.029469977528223976,
      "startedAt": "2026-06-12T20:32:51.000Z",
      "releaseDate": "2025-10-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTFDQUeRKbWiEodRQRrKdpy.eval",
      "sourceId": "TFDQUeRKbWiEodRQRrKdpy",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5-mini-2025-08-07_high",
      "score": 0.4666666666666667,
      "bestScore": 0.4666666666666667,
      "standardError": 0.029603535719125714,
      "startedAt": "2026-06-12T16:48:15.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYSR9WCR2SnJYvP7DJwX7YN.eval",
      "sourceId": "YSR9WCR2SnJYvP7DJwX7YN",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.4-nano-2026-03-17_high",
      "score": 0.44912280701754387,
      "bestScore": 0.44912280701754387,
      "standardError": 0.029515543245521678,
      "startedAt": "2026-06-12T16:48:06.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FiKtopjTjyCWZqJaEjcc7ZL.eval",
      "sourceId": "iKtopjTjyCWZqJaEjcc7ZL",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5-nano-2025-08-07_high",
      "score": 0.2,
      "bestScore": 0.2,
      "standardError": 0.023735633163877067,
      "startedAt": "2026-06-12T16:47:59.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJH3PjDrbrQndEcc8fh5Qre.eval",
      "sourceId": "JH3PjDrbrQndEcc8fh5Qre",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.4-mini-2026-03-17_xhigh",
      "score": 0.512280701754386,
      "bestScore": 0.512280701754386,
      "standardError": 0.02966059084324674,
      "startedAt": "2026-06-12T16:40:38.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "jjFRrQvsud3XsgYufr9f8f",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-2.5-pro",
      "score": 0.24561403508771928,
      "bestScore": 0.24561403508771928,
      "standardError": 0.0255425481026507,
      "startedAt": "2026-06-11T23:56:12.000Z",
      "releaseDate": "2025-06-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFkywQ2KS4AkLPzjSPmhHcS.eval",
      "sourceId": "FkywQ2KS4AkLPzjSPmhHcS",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-3-flash-preview",
      "score": 0.512280701754386,
      "bestScore": 0.512280701754386,
      "standardError": 0.02966059084324674,
      "startedAt": "2026-06-11T23:44:30.000Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FEtgrDq6ei6kZe8K7QbXUe3.eval",
      "sourceId": "EtgrDq6ei6kZe8K7QbXUe3",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-3.1-pro-preview",
      "score": 0.5964912280701754,
      "bestScore": 0.5964912280701754,
      "standardError": 0.02911181956525014,
      "startedAt": "2026-06-11T23:35:57.000Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhVifTUdQZrFEYabgCRu8Qq.eval",
      "sourceId": "hVifTUdQZrFEYabgCRu8Qq",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-opus-4-1-20250805_32K",
      "score": 0.12631578947368421,
      "bestScore": 0.12631578947368421,
      "standardError": 0.0197127354633578,
      "startedAt": "2026-06-11T21:15:07.000Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F5eRkmXCURH2pRbhJoukfnC.eval",
      "sourceId": "5eRkmXCURH2pRbhJoukfnC",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-sonnet-4-5-20250929_32K",
      "score": 0.23859649122807017,
      "bestScore": 0.23859649122807017,
      "standardError": 0.02529183228411416,
      "startedAt": "2026-06-11T21:10:25.000Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7oUkeLfwMK3MtkSNHX6QPu.eval",
      "sourceId": "7oUkeLfwMK3MtkSNHX6QPu",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.4-2026-03-05_xhigh",
      "score": 0.7859649122807018,
      "bestScore": 0.7859649122807018,
      "standardError": 0.024338000553262393,
      "startedAt": "2026-06-11T17:55:45.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPaYEQ37A2n6giYBwZGBnXY.eval",
      "sourceId": "PaYEQ37A2n6giYBwZGBnXY",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.5_xhigh",
      "score": 0.8526315789473684,
      "bestScore": 0.8526315789473684,
      "standardError": 0.021034091168852485,
      "startedAt": "2026-06-11T17:47:36.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5FGXKvXQwGPtcCX2PNrLR5",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o4-mini-2025-04-16_high",
      "score": 0.36140350877192984,
      "bestScore": 0.36140350877192984,
      "standardError": 0.028506918644975,
      "startedAt": "2026-06-11T13:04:19.278Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdcmRi5JTa27EuwiQTdqr2B.eval",
      "sourceId": "dcmRi5JTa27EuwiQTdqr2B",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "o3-mini-2025-01-31_high",
      "score": 0.18596491228070175,
      "bestScore": 0.18596491228070175,
      "standardError": 0.023087552563757593,
      "startedAt": "2026-06-11T13:04:17.668Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmRJaJyBY2M2GGdbdMmT6iq.eval",
      "sourceId": "mRJaJyBY2M2GGdbdMmT6iq",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5.2-2025-12-11_xhigh",
      "score": 0.674,
      "bestScore": 0.674,
      "standardError": 0.028,
      "startedAt": "2026-06-11T02:48:30.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PeZ84GTQdSYQzDxRMWA2Ha",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-opus-4-5-20251101_32K",
      "score": 0.34385964912280703,
      "bestScore": 0.34385964912280703,
      "standardError": 0.028185763989072587,
      "startedAt": "2026-06-11T01:55:25.000Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmS5EwCRZVpvUvJEehPUDZu.eval",
      "sourceId": "mS5EwCRZVpvUvJEehPUDZu",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-opus-4-6_max",
      "score": 0.6596491228070176,
      "bestScore": 0.6596491228070176,
      "standardError": 0.028116467881852295,
      "startedAt": "2026-06-11T00:19:44.000Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FkgJ2d6XbKj3dZutDcBYSKU.eval",
      "sourceId": "kgJ2d6XbKj3dZutDcBYSKU",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-opus-4-7_max",
      "score": 0.7017543859649122,
      "bestScore": 0.7017543859649122,
      "standardError": 0.027146911721220194,
      "startedAt": "2026-06-10T21:41:54.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZkNQ7GER7Mc9FSPmpUzgKH.eval",
      "sourceId": "ZkNQ7GER7Mc9FSPmpUzgKH",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.5543859649122806,
      "bestScore": 0.5543859649122806,
      "standardError": 0.029493504108183504,
      "startedAt": "2026-06-10T20:10:57.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "QRDHTGX3rYrVCjVPRarjjt",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "kimi-k2.6",
      "score": 0.5719298245614035,
      "bestScore": 0.5719298245614035,
      "standardError": 0.029360921879029937,
      "startedAt": "2026-06-10T19:47:36.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FBxrNfyMffkVUq4n7U5eK3Q.eval",
      "sourceId": "BxrNfyMffkVUq4n7U5eK3Q",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "gemini-3.5-flash_high",
      "score": 0.6280701754385964,
      "bestScore": 0.6280701754385964,
      "standardError": 0.028679753751074618,
      "startedAt": "2026-06-10T17:32:38.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F72DQVwbYiQW5pjhshjBaGU.eval",
      "sourceId": "72DQVwbYiQW5pjhshjBaGU",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-opus-4-8_max",
      "score": 0.8,
      "bestScore": 0.8,
      "standardError": 0.023735633163877067,
      "startedAt": "2026-06-10T15:25:15.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGVmgtgKMwEqc54f3jA7auP.eval",
      "sourceId": "GVmgtgKMwEqc54f3jA7auP",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tiers-1-3-v2-Private",
      "model": "claude-fable-5_max",
      "score": 0.8701754385964913,
      "bestScore": 0.8701754385964913,
      "standardError": 0.019944477920133024,
      "startedAt": "2026-06-09T18:12:35.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FE28uZiJvrPgKtawbJpoZQ8.eval",
      "sourceId": "E28uZiJvrPgKtawbJpoZQ8",
      "sourceFile": "frontiermath_tiers_1_3_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "Muse-Spark-Open-816B-A42B_xhigh",
      "score": 0.3170731707317073,
      "bestScore": 0.3170731707317073,
      "standardError": 0.07357611282438223,
      "startedAt": "2026-10-15T18:52:04.000Z",
      "releaseDate": "2026-10-15",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "YifYkfNddMSuE8uBNdbvLi",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gemini-4-argon_high",
      "score": 0.7804878048780488,
      "bestScore": 0.7804878048780488,
      "standardError": 0.06544589202438408,
      "startedAt": "2026-10-13T17:14:57.000Z",
      "releaseDate": "2026-09-30",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6FQCa8tMipf2ncdgGgMaPc",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "mistral-large-4-0_high",
      "score": 0.2682926829268293,
      "bestScore": 0.2682926829268293,
      "standardError": 0.07005564203095158,
      "startedAt": "2026-10-12T21:38:35.000Z",
      "releaseDate": "2026-10-06",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "TzoNRjKto6wEuZ5PmQPAw8",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "muse-wonder-1.0_max",
      "score": 0.7073170731707317,
      "bestScore": 0.7073170731707317,
      "standardError": 0.07194088392074452,
      "startedAt": "2026-10-12T19:00:20.000Z",
      "releaseDate": "2026-10-12",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "76KF9F9owpePuyHpP68mdZ",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "deepseek-v4.1-flash_max",
      "score": 0.2682926829268293,
      "bestScore": 0.2682926829268293,
      "standardError": 0.07005564203095158,
      "startedAt": "2026-10-08T19:02:32.000Z",
      "releaseDate": "2026-09-09",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "KansJcyeGR4m7G4p8GupDn",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-haiku-5-5_max",
      "score": 0.4634146341463415,
      "bestScore": 0.4634146341463415,
      "standardError": 0.07884502378168105,
      "startedAt": "2026-10-07T20:42:31.000Z",
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "WmUXFQoq7GqiSgpTmfLaVv",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-6.1-sol_max",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-09-29T22:00:34.000Z",
      "releaseDate": "2026-09-29",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "HgAUm5zxrWbkfYUdeJMGzL",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-sonnet-5-5_max",
      "score": 0.8048780487804879,
      "bestScore": 0.8048780487804879,
      "standardError": 0.06265967111543964,
      "startedAt": "2026-09-29T15:15:14.000Z",
      "releaseDate": "2026-09-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Xo7ecFEMqPJ4HicvJTvWZZ",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-6-luna_max",
      "score": 0.5609756097560976,
      "bestScore": 0.5609756097560976,
      "standardError": 0.07846686801046543,
      "startedAt": "2026-09-22T22:14:05.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "7n3HBUK684GmdT6Tkycu6m",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "grok-4.7_xhigh",
      "score": 0.17073170731707318,
      "bestScore": 0.17073170731707318,
      "standardError": 0.05949419959829495,
      "startedAt": "2026-09-22T21:49:22.000Z",
      "releaseDate": "2026-09-21",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Ymgj5VeEceBbzsLq6xEZRi",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-6-sol_max",
      "score": 0.9,
      "bestScore": 0.9,
      "standardError": 0.051739520574625435,
      "startedAt": "2026-09-22T19:59:24.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "jwS3LYLCPbZpHDzzkX6LyV",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-opus-5-5_max",
      "score": 0.95,
      "bestScore": 0.95,
      "standardError": 0.041175470771058854,
      "startedAt": "2026-09-22T18:27:39.000Z",
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9qKrTxPhpfW985gieKU56x",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "muse-spark-1.3_max",
      "score": 0.4634146341463415,
      "bestScore": 0.4634146341463415,
      "standardError": 0.07884502378168105,
      "startedAt": "2026-09-18T16:34:54.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "naZno5TPpgMJfv2b5mhvUW",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "muse-spark-1.3_xhigh",
      "score": 0.4146341463414634,
      "bestScore": 0.4146341463414634,
      "standardError": 0.07789619230571372,
      "startedAt": "2026-09-16T02:51:59.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "UdTt275jahRApXURferb9i",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "qwen3.8-max-0902_xhigh",
      "score": 0.34146341463414637,
      "bestScore": 0.34146341463414637,
      "standardError": 0.07497768853141171,
      "startedAt": "2026-09-02T18:04:58.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "kZUdrqdGtCEJa3ip6L9YnH",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gemini-3.8-flash_high",
      "score": 0.21951219512195122,
      "bestScore": 0.21951219512195122,
      "standardError": 0.06544589202438408,
      "startedAt": "2026-09-02T17:15:36.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Lgy52ZT4YpFrUqqKVu2qUs",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-fable-5-1_max",
      "score": 0.8780487804878049,
      "bestScore": 0.8780487804878049,
      "standardError": 0.051739520574625435,
      "startedAt": "2026-09-01T18:38:24.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "T56N3X4KLA3xz7tf5aCMGb",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-6-astra_none",
      "score": 0.8292682926829268,
      "bestScore": 0.8292682926829268,
      "standardError": 0.05949419959829497,
      "startedAt": "2026-08-30T02:22:12.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4BCZTu7HQKmhdwRR2W4bzT",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-6-astra_low",
      "score": 0.8780487804878049,
      "bestScore": 0.8780487804878049,
      "standardError": 0.051739520574625435,
      "startedAt": "2026-08-30T02:22:12.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "kRPSnHnDdheajNdqmKsMjs",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-6-astra_medium",
      "score": 0.975609756097561,
      "bestScore": 0.975609756097561,
      "standardError": 0.024390243902439022,
      "startedAt": "2026-08-30T02:22:12.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5sNRWemuVwb4adBaSwrLFf",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-6-astra_high",
      "score": 0.976,
      "bestScore": 0.976,
      "standardError": 0.024,
      "startedAt": "2026-08-30T02:22:12.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "f46NQY9jSxyKqo9qzrmk6F",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-6-astra_xhigh",
      "score": 0.976,
      "bestScore": 0.976,
      "standardError": 0.024,
      "startedAt": "2026-08-30T02:22:12.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "AfKJ2DGMhkjZmeFZxGoY3w",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-6-astra_max",
      "score": 0.976,
      "bestScore": 0.976,
      "standardError": 0.024,
      "startedAt": "2026-08-30T02:22:12.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "f7BCT27LzcAtaLDP2tcrWF",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "glm-5.3-flash_max",
      "score": 0.17073170731707318,
      "bestScore": 0.17073170731707318,
      "standardError": 0.05949419959829495,
      "startedAt": "2026-08-27T13:29:52.000Z",
      "releaseDate": "2026-08-20",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "iCysWvFJevNyw2dbYosU6J",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "glm-5.3_max",
      "score": 0.2926829268292683,
      "bestScore": 0.2926829268292683,
      "standardError": 0.07194088392074452,
      "startedAt": "2026-08-25T14:20:20.000Z",
      "releaseDate": "2026-08-14",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "4S3ZrJDkwh4gmGrmJQdvzR",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "deepseek-v4-pro-0813_max",
      "score": 0.2682926829268293,
      "bestScore": 0.2682926829268293,
      "standardError": 0.07005564203095158,
      "startedAt": "2026-08-19T13:35:48.000Z",
      "releaseDate": "2026-08-13",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fa4uV8HMMBAcHXLVSmZfBgX.eval",
      "sourceId": "a4uV8HMMBAcHXLVSmZfBgX",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gemini-3.7-flash_high",
      "score": 0.36585365853658536,
      "bestScore": 0.36585365853658536,
      "standardError": 0.07615851217559022,
      "startedAt": "2026-08-14T19:59:58.000Z",
      "releaseDate": "2026-08-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKptKXK5nE2Ns2WbiaSSnbc.eval",
      "sourceId": "KptKXK5nE2Ns2WbiaSSnbc",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "inkling-small_xhigh",
      "score": 0.17073170731707318,
      "bestScore": 0.17073170731707318,
      "standardError": 0.05949419959829495,
      "startedAt": "2026-08-14T19:59:09.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDVD8G8DHNyzkowBzK7JvMt.eval",
      "sourceId": "DVD8G8DHNyzkowBzK7JvMt",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "grok-4.6_xhigh",
      "score": 0.3170731707317073,
      "bestScore": 0.3170731707317073,
      "standardError": 0.07357611282438223,
      "startedAt": "2026-08-14T15:46:50.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fbg3JRdd33iWfvRoGSvyXXE.eval",
      "sourceId": "bg3JRdd33iWfvRoGSvyXXE",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "Inkling_xhigh",
      "score": 0.04878048780487805,
      "bestScore": 0.04878048780487805,
      "standardError": 0.03405912205797303,
      "startedAt": "2026-08-06T20:14:32.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjvJPoRYkPH9Sb5MX252oHp.eval",
      "sourceId": "jvJPoRYkPH9Sb5MX252oHp",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "qwen3.8-max_xhigh",
      "score": 0.4634146341463415,
      "bestScore": 0.4634146341463415,
      "standardError": 0.07884502378168105,
      "startedAt": "2026-08-04T19:32:51.000Z",
      "releaseDate": "2026-08-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPBRdMZyz4L4CqPpXyLD2Lk.eval",
      "sourceId": "PBRdMZyz4L4CqPpXyLD2Lk",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.5-instant",
      "score": 0.024390243902439025,
      "bestScore": 0.024390243902439025,
      "standardError": 0.024390243902439022,
      "startedAt": "2026-08-02T03:23:36.000Z",
      "releaseDate": "2026-05-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2UmuBfhBvYbbnTg5BnQqov.eval",
      "sourceId": "2UmuBfhBvYbbnTg5BnQqov",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "deepseek-v4-flash-0731_max",
      "score": 0.24390243902439024,
      "bestScore": 0.24390243902439024,
      "standardError": 0.06789956540036612,
      "startedAt": "2026-08-02T02:15:37.000Z",
      "releaseDate": "2026-07-31",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWYs83xB6NwrkMNyQJtuMcQ.eval",
      "sourceId": "WYs83xB6NwrkMNyQJtuMcQ",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gemini-3.5-flash-lite_high",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2026-08-02T01:20:17.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjuNVfKBcYTwaKnk9GRA2zL.eval",
      "sourceId": "juNVfKBcYTwaKnk9GRA2zL",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gemini-3.6-flash_high",
      "score": 0.21951219512195122,
      "bestScore": 0.21951219512195122,
      "standardError": 0.06544589202438408,
      "startedAt": "2026-08-02T00:36:45.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJCMjXU6FRiiLV3xvs9R2Xv.eval",
      "sourceId": "JCMjXU6FRiiLV3xvs9R2Xv",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-opus-5_max",
      "score": 0.7317073170731707,
      "bestScore": 0.7317073170731707,
      "standardError": 0.07005564203095158,
      "startedAt": "2026-07-24T19:54:20.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fji3kMn49Jcr9t4whxcpjj4.eval",
      "sourceId": "ji3kMn49Jcr9t4whxcpjj4",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "kimi-k3_max",
      "score": 0.3902439024390244,
      "bestScore": 0.3902439024390244,
      "standardError": 0.07712872341874095,
      "startedAt": "2026-07-17T01:48:42.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgZERSAQwxBAevQzBsWfUie.eval",
      "sourceId": "gZERSAQwxBAevQzBsWfUie",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "grok-4.20-0309-reasoning",
      "score": 0.17073170731707318,
      "bestScore": 0.17073170731707318,
      "standardError": 0.05949419959829495,
      "startedAt": "2026-07-13T15:51:35.000Z",
      "releaseDate": "2026-02-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSX5FCWMk3HF83gGsP3WFA6.eval",
      "sourceId": "SX5FCWMk3HF83gGsP3WFA6",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.6-sol_promax",
      "score": 0.8048780487804879,
      "bestScore": 0.8048780487804879,
      "standardError": 0.06265967111543964,
      "startedAt": "2026-07-09T20:34:41.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcdWwJSH3jVfrKyhUHmT54T.eval",
      "sourceId": "cdWwJSH3jVfrKyhUHmT54T",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.6-terra_max",
      "score": 0.7073170731707317,
      "bestScore": 0.7073170731707317,
      "standardError": 0.07194088392074452,
      "startedAt": "2026-07-09T15:58:30.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZUygpkXqg3NjchXXfedAWV.eval",
      "sourceId": "ZUygpkXqg3NjchXXfedAWV",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.6-luna_max",
      "score": 0.6097560975609756,
      "bestScore": 0.6097560975609756,
      "standardError": 0.07712872341874097,
      "startedAt": "2026-07-09T14:18:35.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FNLiiJJSyA9wmg2FHENFpAq.eval",
      "sourceId": "NLiiJJSyA9wmg2FHENFpAq",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.6-sol_max",
      "score": 0.8292682926829268,
      "bestScore": 0.8292682926829268,
      "standardError": 0.05949419959829497,
      "startedAt": "2026-07-09T02:45:09.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQJ5rJMVypB4PMxPcBGftc8.eval",
      "sourceId": "QJ5rJMVypB4PMxPcBGftc8",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "grok-4.5_high",
      "score": 0.24390243902439024,
      "bestScore": 0.24390243902439024,
      "standardError": 0.06789956540036612,
      "startedAt": "2026-07-09T01:53:42.000Z",
      "releaseDate": "2026-07-08",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "7vThbrtbczScg2UcsAYXtF",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-sonnet-5_max",
      "score": 0.2926829268292683,
      "bestScore": 0.2926829268292683,
      "standardError": 0.07194088392074452,
      "startedAt": "2026-06-30T21:54:08.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbqiVinHXFcgyFGPstJwXJN.eval",
      "sourceId": "bqiVinHXFcgyFGPstJwXJN",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "glm-5.2_max",
      "score": 0.2926829268292683,
      "bestScore": 0.2926829268292683,
      "standardError": 0.07194088392074452,
      "startedAt": "2026-06-19T16:13:05.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJebA8dYLVov6H49ZQSLXh5.eval",
      "sourceId": "JebA8dYLVov6H49ZQSLXh5",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "grok-4.3_high",
      "score": 0.14634146341463414,
      "bestScore": 0.14634146341463414,
      "standardError": 0.055885069450680974,
      "startedAt": "2026-06-17T17:10:48.000Z",
      "releaseDate": "2026-04-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F3iRQkf5v7Wu7YuZhxa76Gg.eval",
      "sourceId": "3iRQkf5v7Wu7YuZhxa76Gg",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "deepseek-v4-pro_max",
      "score": 0.024390243902439025,
      "bestScore": 0.024390243902439025,
      "standardError": 0.024390243902439025,
      "startedAt": "2026-06-17T02:30:14.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRAZ6VMfGtQSE8EvUZRRz4Q.eval",
      "sourceId": "RAZ6VMfGtQSE8EvUZRRz4Q",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "qwen3.7-max",
      "score": 0.34146341463414637,
      "bestScore": 0.34146341463414637,
      "standardError": 0.07497768853141171,
      "startedAt": "2026-06-13T03:08:22.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFHZCKbtfvpKGXU99vZFnXN.eval",
      "sourceId": "FHZCKbtfvpKGXU99vZFnXN",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.4-pro-2026-03-05_xhigh",
      "score": 0.5853658536585366,
      "bestScore": 0.5853658536585366,
      "standardError": 0.07789619230571372,
      "startedAt": "2026-06-13T02:38:02.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FW6b3C8AeNC7CAFs5DQma8N.eval",
      "sourceId": "W6b3C8AeNC7CAFs5DQma8N",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.2-pro-2025-12-11_xhigh",
      "score": 0.46,
      "bestScore": 0.46,
      "standardError": 0.07789619230571372,
      "startedAt": "2026-06-13T02:37:55.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "oALBfwNc7dcNSDJSSBJXYe",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "kimi-k2.7-code",
      "score": 0.12195121951219512,
      "bestScore": 0.12195121951219512,
      "standardError": 0.051739520574625435,
      "startedAt": "2026-06-13T01:39:44.000Z",
      "releaseDate": "2026-06-12",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FUTVSgirxQpTgmTHbgE9EUk.eval",
      "sourceId": "UTVSgirxQpTgmTHbgE9EUk",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.5-pro_xhigh",
      "score": 0.7804878048780488,
      "bestScore": 0.7804878048780488,
      "standardError": 0.06544589202438408,
      "startedAt": "2026-06-12T21:33:12.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "SFKXCPUu73mwxMoUotiZM2",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5-pro-2025-10-06_high",
      "score": 0.1951219512195122,
      "bestScore": 0.1951219512195122,
      "standardError": 0.06265967111543964,
      "startedAt": "2026-06-12T20:32:51.000Z",
      "releaseDate": "2025-10-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FHfDahF4ZXTN6bjfDFwcHTU.eval",
      "sourceId": "HfDahF4ZXTN6bjfDFwcHTU",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5-mini-2025-08-07_high",
      "score": 0.12195121951219512,
      "bestScore": 0.12195121951219512,
      "standardError": 0.051739520574625435,
      "startedAt": "2026-06-12T16:48:15.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2WxRbjq5ZWTgiXZPqYUG4R.eval",
      "sourceId": "2WxRbjq5ZWTgiXZPqYUG4R",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.4-nano-2026-03-17_high",
      "score": 0.12195121951219512,
      "bestScore": 0.12195121951219512,
      "standardError": 0.051739520574625435,
      "startedAt": "2026-06-12T16:48:06.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSF6FYnnKztPCDWQc4v4EJa.eval",
      "sourceId": "SF6FYnnKztPCDWQc4v4EJa",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5-nano-2025-08-07_high",
      "score": 0.024390243902439025,
      "bestScore": 0.024390243902439025,
      "standardError": 0.024390243902439025,
      "startedAt": "2026-06-12T16:47:59.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FP5v3FUnVVTbaqaCr43deKi.eval",
      "sourceId": "P5v3FUnVVTbaqaCr43deKi",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.4-mini-2026-03-17_xhigh",
      "score": 0.0975609756097561,
      "bestScore": 0.0975609756097561,
      "standardError": 0.04691557088212523,
      "startedAt": "2026-06-12T16:40:39.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "jtQ6rDiNxjNKkVMEPBE3Ub",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gdm-ai-co-mathematician",
      "score": 0.756,
      "bestScore": 0.756,
      "standardError": 0.067,
      "startedAt": "2026-06-12T14:01:58.023Z",
      "releaseDate": "2026-05-08",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gdm-ai-co-mathematician-manual-run-v2",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gemini-2.5-pro",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2026-06-11T23:56:12.000Z",
      "releaseDate": "2025-06-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2z7kL9DpfvY44T9Wo99FYx.eval",
      "sourceId": "2z7kL9DpfvY44T9Wo99FYx",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gemini-3-flash-preview",
      "score": 0.17073170731707318,
      "bestScore": 0.17073170731707318,
      "standardError": 0.05949419959829497,
      "startedAt": "2026-06-11T23:44:30.000Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZczEhu69FQ9Cnns8NT6ew4.eval",
      "sourceId": "ZczEhu69FQ9Cnns8NT6ew4",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gemini-3.1-pro-preview",
      "score": 0.2682926829268293,
      "bestScore": 0.2682926829268293,
      "standardError": 0.07005564203095158,
      "startedAt": "2026-06-11T23:35:57.000Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FeXVWr3RBeSsWy5DAU8ed46.eval",
      "sourceId": "eXVWr3RBeSsWy5DAU8ed46",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-opus-4-1-20250805_32K",
      "score": 0.024390243902439025,
      "bestScore": 0.024390243902439025,
      "standardError": 0.024390243902439022,
      "startedAt": "2026-06-11T21:15:07.000Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9zT72xRR6wZhiwraJTNPyz.eval",
      "sourceId": "9zT72xRR6wZhiwraJTNPyz",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-sonnet-4-5-20250929_32K",
      "score": 0.024390243902439025,
      "bestScore": 0.024390243902439025,
      "standardError": 0.024390243902439022,
      "startedAt": "2026-06-11T21:10:25.000Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fd3m9Xip8cof4s77z6LKkEh.eval",
      "sourceId": "d3m9Xip8cof4s77z6LKkEh",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.4-2026-03-05_xhigh",
      "score": 0.49,
      "bestScore": 0.49,
      "standardError": 0.08,
      "startedAt": "2026-06-11T17:55:45.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdoiQDghqnLzbE8CFijLNQJ.eval",
      "sourceId": "doiQDghqnLzbE8CFijLNQJ",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.5_xhigh",
      "score": 0.725,
      "bestScore": 0.725,
      "standardError": 0.07149950690165274,
      "startedAt": "2026-06-11T17:47:36.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PKSMspJiJdNThB6qNvaBSM",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5.2-2025-12-11_xhigh",
      "score": 0.317,
      "bestScore": 0.317,
      "standardError": 0.073,
      "startedAt": "2026-06-11T14:48:04.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "J3NUxsK2GXq9Whkjn7mEVm",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "o3-mini-2025-01-31_high",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2026-06-11T13:04:17.103Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FiR3Ut7aYEtz8dTYQFzq7rG.eval",
      "sourceId": "iR3Ut7aYEtz8dTYQFzq7rG",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "o4-mini-2025-04-16_high",
      "score": 0.04878048780487805,
      "bestScore": 0.04878048780487805,
      "standardError": 0.03405912205797303,
      "startedAt": "2026-06-11T13:02:32.385Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTomkQucNryhFGkaUnUGeLR.eval",
      "sourceId": "TomkQucNryhFGkaUnUGeLR",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-opus-4-5-20251101_32K",
      "score": 0.04878048780487805,
      "bestScore": 0.04878048780487805,
      "standardError": 0.03405912205797303,
      "startedAt": "2026-06-11T01:55:25.000Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fazk4theXxAsizmoi2HDD49.eval",
      "sourceId": "azk4theXxAsizmoi2HDD49",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.21951219512195122,
      "bestScore": 0.21951219512195122,
      "standardError": 0.06544589202438408,
      "startedAt": "2026-06-11T01:44:04.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mSS7C84hJyYi4VnJtgzYvW",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-opus-4-6_max",
      "score": 0.2682926829268293,
      "bestScore": 0.2682926829268293,
      "standardError": 0.07005564203095158,
      "startedAt": "2026-06-11T00:19:44.000Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2hbbp8z6PhB7c2T33os9q5.eval",
      "sourceId": "2hbbp8z6PhB7c2T33os9q5",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-opus-4-7_max",
      "score": 0.3170731707317073,
      "bestScore": 0.3170731707317073,
      "standardError": 0.07357611282438223,
      "startedAt": "2026-06-10T21:41:54.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FP4LXQ9PyDDziEANMB5jVRd.eval",
      "sourceId": "P4LXQ9PyDDziEANMB5jVRd",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "kimi-k2.6",
      "score": 0.2564102564102564,
      "bestScore": 0.2564102564102564,
      "standardError": 0.07083413480167725,
      "startedAt": "2026-06-10T19:47:36.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FeVKA2eavdRAbSN6AnzyBxA.eval",
      "sourceId": "eVKA2eavdRAbSN6AnzyBxA",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "gemini-3.5-flash_high",
      "score": 0.2682926829268293,
      "bestScore": 0.2682926829268293,
      "standardError": 0.07005564203095158,
      "startedAt": "2026-06-10T17:32:38.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcXn5zzns6VFwbTgotDwZjV.eval",
      "sourceId": "cXn5zzns6VFwbTgotDwZjV",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-opus-4-8_max",
      "score": 0.5609756097560976,
      "bestScore": 0.5609756097560976,
      "standardError": 0.07846686801046543,
      "startedAt": "2026-06-10T15:25:15.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJiqgqBTfkGx797uowZgJbk.eval",
      "sourceId": "JiqgqBTfkGx797uowZgJbk",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "FrontierMath-Tier-4-v2-Private",
      "model": "claude-fable-5_max",
      "score": 0.902,
      "bestScore": 0.902,
      "standardError": 0.046,
      "startedAt": "2026-06-09T18:12:35.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FReDAtJGGfRtrEGGWf4Zbsd.eval",
      "sourceId": "ReDAtJGGfRtrEGGWf4Zbsd",
      "sourceFile": "frontiermath_tier_4_v2.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-5-mini-2025-08-07_high",
      "score": 0.9784743202416919,
      "bestScore": 0.9784743202416919,
      "standardError": 0.003369422414145465,
      "startedAt": "2025-10-30T10:00:02.998Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5duoiNDccsgaW7UVt6feYX",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.9813066465256798,
      "bestScore": 0.9813066465256798,
      "standardError": 0.0031922231523012308,
      "startedAt": "2025-10-29T22:29:43.092Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4wN6FKe2db9CLQ3NUFfYge",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-haiku-4-5-20251001_32K",
      "score": 0.9636087945413192,
      "bestScore": 0.9636087945413192,
      "standardError": 0.00515811348923114,
      "startedAt": "2025-10-22T13:37:59.284Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Qwci6JAG4DUWqA9eSHThDJ",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-sonnet-4-5-20250929_32K",
      "score": 0.9773413897280967,
      "bestScore": 0.9773413897280967,
      "standardError": 0.004091286174812221,
      "startedAt": "2025-10-21T16:28:50.811Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "FqjAH25VNjgwQTbDJFfqAg",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-haiku-4-5-20251001",
      "score": 0.8691465256797583,
      "bestScore": 0.8691465256797583,
      "standardError": 0.007747551687214772,
      "startedAt": "2025-10-16T11:58:16.245Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "3w2dD9rcMPHLYsEsJmYsCX",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "qwen3-max-2025-09-23",
      "score": 0.9712990936555891,
      "bestScore": 0.9712990936555891,
      "standardError": 0.004200694815744905,
      "startedAt": "2025-10-09T15:30:36.002Z",
      "releaseDate": "2025-09-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "f2vxirYMpXwkaZxb4ywnYM",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-5-2025-08-07_medium",
      "score": 0.9792296072507553,
      "bestScore": 0.9792296072507553,
      "standardError": 0.003446256264439923,
      "startedAt": "2025-08-20T11:28:53.620Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "eDzsqj2CnVxbkp5JRuULXF",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-5-nano-2025-08-07_medium",
      "score": 0.952416918429003,
      "bestScore": 0.952416918429003,
      "standardError": 0.004627056893049539,
      "startedAt": "2025-08-20T11:28:52.351Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "SSLfQCm7RvLhSv5aPKaLNk",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-5-nano-2025-08-07_high",
      "score": 0.9490181268882175,
      "bestScore": 0.9490181268882175,
      "standardError": 0.004848851575113763,
      "startedAt": "2025-08-20T11:28:52.016Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "TYQuXJw9H8NRjSovr5VoBn",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-5-mini-2025-08-07_medium",
      "score": 0.9679003021148036,
      "bestScore": 0.9679003021148036,
      "standardError": 0.003853475747353661,
      "startedAt": "2025-08-20T11:28:51.670Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "3NYyCGnM6XhgytjyVGtwh8",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "qwen3-235b-a22b",
      "score": 0.6885699899295065,
      "bestScore": 0.6885699899295065,
      "standardError": 0.009487067376018314,
      "startedAt": "2025-06-03T22:33:05.178Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "BTDXT8mHQu4je5jysTDuXV",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "DeepSeek-R1-0528",
      "score": 0.9663897280966768,
      "bestScore": 0.9663897280966768,
      "standardError": 0.0042156161735636515,
      "startedAt": "2025-05-29T14:09:30.970Z",
      "releaseDate": "2025-05-28",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "2wd9khVKiBzsAMUquz2R5s",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-sonnet-4-20250514",
      "score": 0.8436555891238671,
      "bestScore": 0.8436555891238671,
      "standardError": 0.00998490304339695,
      "startedAt": "2025-05-22T17:39:31.972Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dZLNhNKQVUgnB5ygAEUCcT",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-opus-4-20250514",
      "score": 0.850453172205438,
      "bestScore": 0.850453172205438,
      "standardError": 0.009804690367660946,
      "startedAt": "2025-05-22T17:39:30.135Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "YbrpPbWuYAXNqbSfU4so7f",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemini-2.5-pro-preview-05-06",
      "score": 0.9590256797583081,
      "bestScore": 0.9590256797583081,
      "standardError": 0.004221388011004008,
      "startedAt": "2025-05-08T05:28:17.657Z",
      "releaseDate": "2025-05-06",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Mi8LCxmduKukxazAtmuhqg",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemini-2.5-pro-preview-03-25",
      "score": 0.9556268882175226,
      "bestScore": 0.9556268882175226,
      "standardError": 0.004475169260814502,
      "startedAt": "2025-05-07T21:59:53.901Z",
      "releaseDate": "2025-03-31",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XRkZCzSoQC5yYiQ3xitNdC",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "mistral-medium-2505",
      "score": 0.8162764350453172,
      "bestScore": 0.8162764350453172,
      "standardError": 0.00899565165100951,
      "startedAt": "2025-05-07T17:32:19.064Z",
      "releaseDate": "2025-05-07",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "LNUnvvrDra89Vhdc9bzdCy",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "o4-mini-2025-04-16_high",
      "score": 0.978285498489426,
      "bestScore": 0.978285498489426,
      "standardError": 0.0033204985518619605,
      "startedAt": "2025-04-16T18:17:08.504Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "727ixGKZ7A7G8LmavE2ScL",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "o3-2025-04-16_high",
      "score": 0.9777190332326284,
      "bestScore": 0.9777190332326284,
      "standardError": 0.0033230338273894273,
      "startedAt": "2025-04-16T18:17:06.429Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6oW3GceNQznewQJRwLigth",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4.1-mini-2025-04-14",
      "score": 0.8729229607250756,
      "bestScore": 0.8729229607250756,
      "standardError": 0.007298484758774334,
      "startedAt": "2025-04-14T22:50:03.233Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6ybVweWdTfiCuirnS2BKXD",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4.1-nano-2025-04-14",
      "score": 0.6999622356495468,
      "bestScore": 0.6999622356495468,
      "standardError": 0.010231591350302844,
      "startedAt": "2025-04-14T22:49:56.785Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "K4TfMXsaPd4Z46zFmiV4Rf",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4.1-2025-04-14",
      "score": 0.8300604229607251,
      "bestScore": 0.8300604229607251,
      "standardError": 0.008490146044318322,
      "startedAt": "2025-04-14T22:22:31.727Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "YyJXPVXJjN5EnLfYjfCTrQ",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "grok-3-mini-beta_low",
      "score": 0.9093655589123867,
      "bestScore": 0.9093655589123867,
      "standardError": 0.007892888346623621,
      "startedAt": "2025-04-10T20:30:31.009Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5QyCm49Lce2PpGVcZLomCz",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "grok-3-mini-beta_high",
      "score": 0.8806646525679759,
      "bestScore": 0.8806646525679759,
      "standardError": 0.008912712785411623,
      "startedAt": "2025-04-10T18:02:30.275Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "HMm8XaDBr4CQdMFddasEWb",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "grok-3-beta",
      "score": 0.8874622356495468,
      "bestScore": 0.8874622356495468,
      "standardError": 0.008688487093324447,
      "startedAt": "2025-04-10T17:55:22.706Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "i9eqMbaVTDDuCxWxzGHxyU",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Llama-4-Maverick-17B-128E-Instruct-FP8",
      "score": 0.7301737160120846,
      "bestScore": 0.7301737160120846,
      "standardError": 0.01083573699636281,
      "startedAt": "2025-04-08T09:46:43.408Z",
      "releaseDate": "2025-04-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "HkjCosgMFq5QXB2CyTyU7r",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Llama-4-Scout-17B-16E-Instruct",
      "score": 0.6227341389728097,
      "bestScore": 0.6227341389728097,
      "standardError": 0.012022405873783983,
      "startedAt": "2025-04-08T09:22:06.630Z",
      "releaseDate": "2025-04-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "c6rTyTFCeoHMBzC2QwjxFL",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "qwen-turbo-2024-11-01",
      "score": 0.5623111782477341,
      "bestScore": 0.5623111782477341,
      "standardError": 0.011364630947550552,
      "startedAt": "2025-04-07T19:55:58.553Z",
      "releaseDate": "2024-11-01",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "jWV936hYKg6rYcPEynzEEt",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "qwen-plus-2025-01-25",
      "score": 0.6527567975830816,
      "bestScore": 0.6527567975830816,
      "standardError": 0.011334654774046507,
      "startedAt": "2025-04-07T17:38:39.334Z",
      "releaseDate": "2025-01-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "J3WoEpLKAJZWGGpyBVCCCn",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "qwen-max-2025-01-25",
      "score": 0.6718277945619335,
      "bestScore": 0.6718277945619335,
      "standardError": 0.011143967442882196,
      "startedAt": "2025-04-01T17:50:34.087Z",
      "releaseDate": "2025-01-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "GFZDNH95TnVegNrEtmUT2Q",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "DeepSeek-V3-0324",
      "score": 0.75547583081571,
      "bestScore": 0.75547583081571,
      "standardError": 0.008603173106577204,
      "startedAt": "2025-04-01T14:52:18.473Z",
      "releaseDate": "2025-03-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "YFyn28VJ6m2hfop3KGRkBs",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "mistral-small-2503",
      "score": 0.4677114803625378,
      "bestScore": 0.4677114803625378,
      "standardError": 0.011995700377965737,
      "startedAt": "2025-03-18T16:08:03.120Z",
      "releaseDate": "2025-03-17",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "Kas7Y5PBsT7Li5Wm6Y8mj8",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-3-7-sonnet-20250219_64K",
      "score": 0.9116314199395771,
      "bestScore": 0.9116314199395771,
      "standardError": 0.007803306377082018,
      "startedAt": "2025-03-13T21:33:10.936Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "epCyFNTf6vyUi5Dby7GeTq",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemma-3-27b-it",
      "score": 0.740370090634441,
      "bestScore": 0.740370090634441,
      "standardError": 0.011195857525727288,
      "startedAt": "2025-03-13T05:52:35.397Z",
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "amrtN7UjAtuyEmv5DaGshP",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-3-5-haiku-20241022",
      "score": 0.46355740181268884,
      "bestScore": 0.46355740181268884,
      "standardError": 0.011611728105526228,
      "startedAt": "2025-03-12T23:51:45.093Z",
      "releaseDate": "2024-10-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GSvi9Zf5aK3t5TfCPqNRHy",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-3-7-sonnet-20250219_32K",
      "score": 0.9003021148036254,
      "bestScore": 0.9003021148036254,
      "standardError": 0.008236774684678254,
      "startedAt": "2025-03-12T19:45:08.221Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "WHGMHrneqJs4nefAh4TjZR",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "DeepSeek-R1-Distill-Qwen-14B",
      "score": 0.8712235649546828,
      "bestScore": 0.8712235649546828,
      "standardError": 0.007486217204155208,
      "startedAt": "2025-03-11T00:50:44.877Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "R8XGt5ScHpmTbD5DnXR59V",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "DeepSeek-R1-Distill-Llama-70B",
      "score": 0.8989803625377644,
      "bestScore": 0.8989803625377644,
      "standardError": 0.006818015200174493,
      "startedAt": "2025-03-10T22:53:10.180Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "oPDoTJFMhMNvU4sybYu2jQ",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4.5-preview-2025-02-27",
      "score": 0.7862537764350453,
      "bestScore": 0.7862537764350453,
      "standardError": 0.011270694428136297,
      "startedAt": "2025-02-28T16:37:00.565Z",
      "releaseDate": "2025-02-27",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "japqZFi3FS7um32EHG6wwW",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4-turbo-2024-04-09",
      "score": 0.46733383685800606,
      "bestScore": 0.46733383685800606,
      "standardError": 0.01142825893039598,
      "startedAt": "2025-02-27T14:29:46.292Z",
      "releaseDate": "2024-04-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "AZAJomaSEfZtqYqXUre9fS",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-3-7-sonnet-20250219_16K",
      "score": 0.8625377643504532,
      "bestScore": 0.8625377643504532,
      "standardError": 0.009466748483397175,
      "startedAt": "2025-02-26T16:56:42.885Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hWeVyRo3hN5H4AspNc9cuf",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "mistral-large-2411",
      "score": 0.5028323262839879,
      "bestScore": 0.5028323262839879,
      "standardError": 0.011469451623480956,
      "startedAt": "2025-02-25T12:31:34.494Z",
      "releaseDate": "2024-11-18",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "MMqL6fQSwsBKAdBSfxrYpU",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-3-7-sonnet-20250219",
      "score": 0.6818353474320241,
      "bestScore": 0.6818353474320241,
      "standardError": 0.010669651861930321,
      "startedAt": "2025-02-24T19:14:03.027Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6uSKG2QxR8atueXmDWAPXn",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "grok-2-1212",
      "score": 0.6351963746223565,
      "bestScore": 0.6351963746223565,
      "standardError": 0.01323438169590836,
      "startedAt": "2025-02-17T21:39:16.270Z",
      "releaseDate": "2024-12-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Kpck3pLw3veM5Mt7WZAb8G",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "o1-2024-12-17_high",
      "score": 0.947129909365559,
      "bestScore": 0.947129909365559,
      "standardError": 0.006152192005965586,
      "startedAt": "2025-02-13T17:12:08.885Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "jVXtRkTBDnqH8p4sjVUbgN",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "o1-mini-2024-09-12_high",
      "score": 0.8918051359516617,
      "bestScore": 0.8918051359516617,
      "standardError": 0.006802820166065648,
      "startedAt": "2025-02-13T17:07:52.489Z",
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4DWnpJowAfeQDHc2xDNy2C",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "o3-mini-2025-01-31_high",
      "score": 0.9648791540785498,
      "bestScore": 0.9648791540785498,
      "standardError": 0.004265263461984669,
      "startedAt": "2025-02-13T17:07:49.918Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "3VDhPT5WvzjJZkRc36Kuvi",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemini-2.0-pro-exp-02-05",
      "score": 0.8345921450151057,
      "bestScore": 0.8345921450151057,
      "standardError": 0.01021492751504924,
      "startedAt": "2025-02-06T17:07:01.731Z",
      "releaseDate": "2025-02-05",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "fvDt6HaHV5d3gSEuavixsv",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemini-2.0-flash-001",
      "score": 0.8216578549848943,
      "bestScore": 0.8216578549848943,
      "standardError": 0.008734081505176936,
      "startedAt": "2025-02-06T15:51:21.031Z",
      "releaseDate": "2025-02-05",
      "organization": "Google DeepMind,Google",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "SqwGZz2UfTmbvPiw2rq3LH",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4o-2024-11-20",
      "score": 0.49773413897280966,
      "bestScore": 0.49773413897280966,
      "standardError": 0.011102819618096572,
      "startedAt": "2025-02-05T08:31:16.384Z",
      "releaseDate": "2024-11-20",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "D3G2BKGnh6YSvw4bEYSdiY",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Phi-3-medium-128k-instruct",
      "score": 0.17560422960725075,
      "bestScore": 0.17560422960725075,
      "standardError": 0.007295657661098519,
      "startedAt": "2025-01-31T23:48:20.023Z",
      "releaseDate": "2024-04-23",
      "organization": "Microsoft",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "njA86GC9DzSXyiZWtUxXiD",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "DeepSeek-R1",
      "score": 0.9305135951661632,
      "bestScore": 0.9305135951661632,
      "standardError": 0.006990872160031429,
      "startedAt": "2025-01-31T23:48:19.611Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "XVqLs2khGFSNwEbHJTEgRU",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "phi-4",
      "score": 0.6493580060422961,
      "bestScore": 0.6493580060422961,
      "standardError": 0.010656871384084826,
      "startedAt": "2025-01-31T23:48:19.147Z",
      "releaseDate": "2024-12-12",
      "organization": "Microsoft Research",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PnvmfAHVxH9oXfcdcUeUZ3",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "o3-mini-2025-01-31_medium",
      "score": 0.9516616314199395,
      "bestScore": 0.9516616314199395,
      "standardError": 0.004651917857621578,
      "startedAt": "2025-01-31T19:32:36.412Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "2RiSB2ii5kf2Q6fDBv2P3E",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "qwen2.5-32b-instruct",
      "score": 0.5607061933534743,
      "bestScore": 0.5607061933534743,
      "standardError": 0.01076720798872177,
      "startedAt": "2025-01-30T19:02:48.179Z",
      "releaseDate": "2024-09-17",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "Nk4mmCXFH3mEyCYnQPdPx5",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "mistral-small-2501",
      "score": 0.4481684290030212,
      "bestScore": 0.4481684290030212,
      "standardError": 0.010854927233993669,
      "startedAt": "2025-01-30T17:26:30.440Z",
      "releaseDate": "2025-01-25",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "NbsnvBsMoMizozbPZY8LLb",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4-0613",
      "score": 0.22970166163141995,
      "bestScore": 0.22970166163141995,
      "standardError": 0.008465534027077483,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-06-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "WsSrH8oHBm7m7vq9yTRqSC",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemini-1.5-flash-001",
      "score": 0.2512273413897281,
      "bestScore": 0.2512273413897281,
      "standardError": 0.009456393246526343,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-05-23",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "CPpd4hqQHtZ5pUejBELwKb",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemini-1.5-pro-002",
      "score": 0.7039274924471299,
      "bestScore": 0.7039274924471299,
      "standardError": 0.010174893415515739,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "QyqLmgcJpeDMsnQyaaJVkR",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-3.5-turbo-0125",
      "score": 0.1163141993957704,
      "bestScore": 0.1163141993957704,
      "standardError": 0.005543945913851348,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-01-25",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9tSgxGV3nzhTBpq7GcX7yx",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemini-1.5-pro-001",
      "score": 0.4074773413897281,
      "bestScore": 0.4074773413897281,
      "standardError": 0.010574902459145802,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-05-14",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PxLmyq7PFfARAuqxg5oPtE",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4-1106-preview",
      "score": 0.40020770392749244,
      "bestScore": 0.40020770392749244,
      "standardError": 0.010538794632397387,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-11-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "VBStRVGFi47EVz2pn33xu4",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-3.5-turbo-1106",
      "score": 0.15889350453172205,
      "bestScore": 0.15889350453172205,
      "standardError": 0.006959408516001194,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-11-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "cJWh887zSiTtnxm9XAJVkr",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemini-1.5-flash-002",
      "score": 0.6186744712990937,
      "bestScore": 0.6186744712990937,
      "standardError": 0.01120122127580226,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "kDnfjYKQpnLGEjuhBc9AvM",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-2.0",
      "score": 0.1172583081570997,
      "bestScore": 0.1172583081570997,
      "standardError": 0.006595670913969979,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-07-11",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Gs4vxvDD5wmrSvyTkqx2YV",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-3-5-sonnet-20241022",
      "score": 0.5694864048338368,
      "bestScore": 0.5694864048338368,
      "standardError": 0.011091284283577038,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-10-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "AjroyuZECXLtMTwnzaMHtX",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-3-5-sonnet-20240620",
      "score": 0.5168051359516617,
      "bestScore": 0.5168051359516617,
      "standardError": 0.0116250120233614,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-06-20",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "H42HJG29bwqzaMewTMsmky",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-3-opus-20240229",
      "score": 0.37481117824773413,
      "bestScore": 0.37481117824773413,
      "standardError": 0.01058728426376393,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-02-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6SNzvpd6kXXxqpJoaHsSDd",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-3-haiku-20240307",
      "score": 0.1487915407854985,
      "bestScore": 0.1487915407854985,
      "standardError": 0.007229965913719499,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-03-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "QSnWVKUVj9six4jYj4HJD9",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "claude-3-sonnet-20240229",
      "score": 0.18174093655589124,
      "bestScore": 0.18174093655589124,
      "standardError": 0.007981817344047291,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-02-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "FD2ztfjFUeqGyrtF8yuCUB",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Llama-3.1-8B-Instruct",
      "score": 0.22875755287009064,
      "bestScore": 0.22875755287009064,
      "standardError": 0.008519457631995396,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gyxR3YTcmh4UcKZkET5pPX",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Llama-3.1-405B-Instruct",
      "score": 0.49773413897280966,
      "bestScore": 0.49773413897280966,
      "standardError": 0.012185275936569536,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZR7amdUBRr5Cke8Aoz4snr",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Llama-3.1-70B-Instruct",
      "score": 0.36678625377643503,
      "bestScore": 0.36678625377643503,
      "standardError": 0.010229604859690931,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6JiXiQ67iyiC7cFFRiogvs",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4o-2024-05-13",
      "score": 0.5104796072507553,
      "bestScore": 0.5104796072507553,
      "standardError": 0.01120841984218505,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-05-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Uqsvq9TAGoqD2xpKfhHBmS",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4o-mini-2024-07-18",
      "score": 0.5263406344410876,
      "bestScore": 0.5263406344410876,
      "standardError": 0.011275275696871733,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-07-18",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5iyKgWVhAmNwrA5rNL7MbW",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4o-2024-08-06",
      "score": 0.5327605740181269,
      "bestScore": 0.5327605740181269,
      "standardError": 0.01122129021441016,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-08-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZQgz8AGm4KHin9MFupLu58",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "o1-mini-2024-09-12_medium",
      "score": 0.8425226586102719,
      "bestScore": 0.8425226586102719,
      "standardError": 0.008256715083788129,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "FaRZDKSriUbS3Brt6qprP9",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Hermes-2-Theta-Llama-3-70B",
      "score": 0.22686933534743203,
      "bestScore": 0.22686933534743203,
      "standardError": 0.008127708089237323,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-06-20",
      "organization": "Nous Research,Arcee AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZAVTwsDqrvbWbqSgeNdXFd",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Meta-Llama-3-8B-Instruct",
      "score": 0.0612726586102719,
      "bestScore": 0.0612726586102719,
      "standardError": 0.0037433164863856527,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-04-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "BURdKU7mhUX4YMPUDoFsed",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Meta-Llama-3-70B-Instruct",
      "score": 0.225547583081571,
      "bestScore": 0.225547583081571,
      "standardError": 0.008257639939780867,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-04-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dekbZbU3BtiQZNKG8Txaax",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Llama-2-70b-chat-hf",
      "score": 0.032854984894259816,
      "bestScore": 0.032854984894259816,
      "standardError": 0.00254222977415661,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-07-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4bkXhqCNoe6RpTe9taJ7kR",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Yi-1.5-34B-Chat",
      "score": 0.2548149546827795,
      "bestScore": 0.2548149546827795,
      "standardError": 0.008772368987739622,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-05-13",
      "organization": "01.AI",
      "country": "China",
      "logViewer": null,
      "sourceId": "bH5kVfLE53CxY3qWmB9C5K",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Yi-34B-Chat",
      "score": 0.05145392749244713,
      "bestScore": 0.05145392749244713,
      "standardError": 0.0033693214361751756,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-11-22",
      "organization": "01.AI",
      "country": "China",
      "logViewer": null,
      "sourceId": "nzy4SGSeKGzQSKJLWExKpV",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gpt-4-0125-preview",
      "score": 0.3541351963746224,
      "bestScore": 0.3541351963746224,
      "standardError": 0.010855237562085883,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-01-25",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "c4V7sc8TUNJ8wQKRhuA2Rv",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemma-2-9b-it",
      "score": 0.2100641993957704,
      "bestScore": 0.2100641993957704,
      "standardError": 0.008990724846256171,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-06-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Hdcqs42VnPNVBquXttexhq",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemma-2-27b-it",
      "score": 0.2788897280966767,
      "bestScore": 0.2788897280966767,
      "standardError": 0.009939474306804167,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-06-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PM3GCQcVK7t2MujNp4F4kH",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "qwen2-72b-instruct",
      "score": 0.39067220543806647,
      "bestScore": 0.39067220543806647,
      "standardError": 0.010317258861464411,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-06-07",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "9mWoY9fM2ifKvDg3xaug6h",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Mistral-7B-Instruct-v0.3",
      "score": 0.03597054380664653,
      "bestScore": 0.03597054380664653,
      "standardError": 0.002666943446495743,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-05-27",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "6xHApTdQAWM9e5oCvKYS3W",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Mixtral-8x7B-Instruct-v0.1",
      "score": 0.09290030211480363,
      "bestScore": 0.09290030211480363,
      "standardError": 0.005102464546317853,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-12-11",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "4PTATufH2ra3guNeLwT9p9",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "deepseek-llm-67b-chat",
      "score": 0.06391616314199396,
      "bestScore": 0.06391616314199396,
      "standardError": 0.004016844149550666,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-11-29",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "Cm4UkgA5TtmzQegrq3rbvm",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "WizardLM-2-8x22B",
      "score": 0.25736404833836857,
      "bestScore": 0.25736404833836857,
      "standardError": 0.008719521575190186,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-04-15",
      "organization": "Microsoft",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4bAYQuh4heJoTNhapVZLNB",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "dbrx-instruct",
      "score": 0.11650302114803625,
      "bestScore": 0.11650302114803625,
      "standardError": 0.007355853708151387,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-03-27",
      "organization": "Databricks",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PMbefrjykQmTBpABMa7FJi",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "gemini-1.0-pro-001",
      "score": 0.11244335347432025,
      "bestScore": 0.11244335347432025,
      "standardError": 0.005990949270731778,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-12-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iUdecHq9vU5utytzgTXzqY",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "ministral-3b-2410",
      "score": 0.14444864048338368,
      "bestScore": 0.14444864048338368,
      "standardError": 0.007446287617112978,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-10-16",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "YNhHxRT9Ee2uEWWJh9RXuP",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "open-mistral-7b",
      "score": 0.0368202416918429,
      "bestScore": 0.0368202416918429,
      "standardError": 0.0027715208474754028,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-09-27",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "9ZQaypWLwLHbQrqTWfoEor",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "ministral-8b-2410",
      "score": 0.14935800604229607,
      "bestScore": 0.14935800604229607,
      "standardError": 0.007454534670584494,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-10-16",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "hqdiKcvEPSRyvenUVB9qxM",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "mistral-large-2407",
      "score": 0.4481684290030212,
      "bestScore": 0.4481684290030212,
      "standardError": 0.010703488686415494,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-07-24",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "div9uE9eovqykqx883kSpE",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "open-mistral-nemo-2407",
      "score": 0.1082892749244713,
      "bestScore": 0.1082892749244713,
      "standardError": 0.006092598712054613,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-07-18",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "9qCrggpoq8FjQWVQhHaGBT",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "open-mixtral-8x7b",
      "score": 0.09950906344410876,
      "bestScore": 0.09950906344410876,
      "standardError": 0.005303272245515919,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2023-12-11",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "GN3Hez798byishFw3U6cpp",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "open-mixtral-8x22b",
      "score": 0.24244712990936557,
      "bestScore": 0.24244712990936557,
      "standardError": 0.008296884112232158,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-04-17",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "cGMb2vs9NhfkwZJsQXzLmQ",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "mistral-large-2402",
      "score": 0.24461858006042297,
      "bestScore": 0.24461858006042297,
      "standardError": 0.00892334786643176,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-02-26",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "gypQecfFNK2x4pNGDtGMDv",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Llama-3.2-90B-Vision-Instruct",
      "score": 0.39435422960725075,
      "bestScore": 0.39435422960725075,
      "standardError": 0.010293053094701802,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-24",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iHRmwxsh8M3FGun8x5mxkt",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Llama-3.1-Tulu-3-70B-DPO",
      "score": 0.426642749244713,
      "bestScore": 0.426642749244713,
      "standardError": 0.010405854683575872,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-11-21",
      "organization": "Allen Institute for AI,University of Washington",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Hf2ESooSteQK99yFjhFm8t",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "Llama-3.3-70B-Instruct",
      "score": 0.41597432024169184,
      "bestScore": 0.41597432024169184,
      "standardError": 0.011816670264886683,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-12-06",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "DeaDxFfiQtCTdnZPH7Awaf",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "qwen2.5-72b-instruct",
      "score": 0.631703172205438,
      "bestScore": 0.631703172205438,
      "standardError": 0.01071424933114761,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "ni4DJboYYpdRssVeAuro2V",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "o1-preview-2024-09-12",
      "score": 0.8164652567975831,
      "bestScore": 0.8164652567975831,
      "standardError": 0.010642608258161584,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8Uu6XsSJCKwcwhXDt8vgzq",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "o1-2024-12-17_medium",
      "score": 0.9441087613293051,
      "bestScore": 0.9441087613293051,
      "standardError": 0.006315430485333487,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "AzanNoyhhVrtQi63Azo6hu",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "MATH level 5",
      "model": "DeepSeek-V3",
      "score": 0.6485083081570997,
      "bestScore": 0.6485083081570997,
      "standardError": 0.010485766934209063,
      "startedAt": "2025-01-27T00:00:00.000Z",
      "releaseDate": "2024-12-26",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "exvQCLCiga6gE7zvQNp2Z6",
      "sourceFile": "math_level_5.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Muse-Spark-Open-816B-A42B_xhigh",
      "score": 0.9805555555555555,
      "bestScore": 0.9805555555555555,
      "standardError": 0.01253502388483925,
      "startedAt": "2026-10-15T18:52:04.000Z",
      "releaseDate": "2026-10-15",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hB4XzBdWkdpG3WJVYqeRM8",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-4-argon_high",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-10-13T17:14:57.000Z",
      "releaseDate": "2026-09-30",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "43f36RTDoJGxXWQDVvTLJj",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "mistral-large-4-0_high",
      "score": 0.9833333333333333,
      "bestScore": 0.9833333333333333,
      "standardError": 0.006405884289881728,
      "startedAt": "2026-10-12T21:38:35.000Z",
      "releaseDate": "2026-10-06",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "Pby5WuP8Muu6Jzs2v5gB7F",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "muse-wonder-1.0_max",
      "score": 0.9944444444444445,
      "bestScore": 0.9944444444444445,
      "standardError": 0.003883473863441843,
      "startedAt": "2026-10-12T19:00:20.000Z",
      "releaseDate": "2026-10-12",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "AFigcTGzpgZgMi6j2fpboa",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "deepseek-v4.1-flash_max",
      "score": 0.9833333333333333,
      "bestScore": 0.9833333333333333,
      "standardError": 0.006405884289881728,
      "startedAt": "2026-10-08T19:02:32.000Z",
      "releaseDate": "2026-09-09",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "7eHNiVQCzLcAyj5LxmeeSU",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-haiku-5-5_xhigh",
      "score": 0.9888888888888889,
      "bestScore": 0.9888888888888889,
      "standardError": 0.005362818328685681,
      "startedAt": "2026-10-07T21:16:36.000Z",
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iiEPrSjWXHH6Zq9CK4Zmcr",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-haiku-5-5_high",
      "score": 0.9722222222222222,
      "bestScore": 0.9722222222222222,
      "standardError": 0.013109441770118595,
      "startedAt": "2026-10-07T21:16:25.000Z",
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LycbwGumF6Q7ySpmGY94XW",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-6.1-sol_max",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-09-29T22:00:34.000Z",
      "releaseDate": "2026-09-29",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Di7y6sES4PGQbXNkXYhdqM",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-5-5_max",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-09-29T15:15:14.000Z",
      "releaseDate": "2026-09-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bSEaZLPehiNVDQL7HP9EJS",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-6-luna_max",
      "score": 0.9888888888888889,
      "bestScore": 0.9888888888888889,
      "standardError": 0.006674027922425511,
      "startedAt": "2026-09-22T22:14:05.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4uDz9b6SosxDSZf7swMD9D",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "grok-4.7_xhigh",
      "score": 0.9805555555555555,
      "bestScore": 0.9805555555555555,
      "standardError": 0.014299486306926387,
      "startedAt": "2026-09-22T21:49:22.000Z",
      "releaseDate": "2026-09-21",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Jhgwm7QpspGcvm7DUVqUDg",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-6-sol_max",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-09-22T19:59:24.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "VZAUcNkzeMSMm9paxBhj6B",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-5-5_max",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-09-22T18:27:39.000Z",
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LSfona7P3jCAnZFXJ2YA7F",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "muse-spark-1.3_xhigh",
      "score": 0.9916666666666667,
      "bestScore": 0.9916666666666667,
      "standardError": 0.006154574548966636,
      "startedAt": "2026-09-17T13:30:27.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Wmxkx5m3qaQuwzem6gzZmg",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.8-max-0902_xhigh",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-09-02T18:04:58.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "2f8Vn5QBdJnc9i5y9jzXS7",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.8-flash_high",
      "score": 0.9888888888888889,
      "bestScore": 0.9888888888888889,
      "standardError": 0.008724007409153888,
      "startedAt": "2026-09-02T17:15:36.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "NeYuqgQyWtFWAHEjQYPm5Z",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-fable-5-1_max",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-09-01T18:32:00.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mRAASXsUuoJmc2oR5G89zR",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "granite-4.0-micro",
      "score": 0.027777777777777776,
      "bestScore": 0.027777777777777776,
      "standardError": 0.014262647723679736,
      "startedAt": "2026-08-30T22:47:13.000Z",
      "releaseDate": "2025-10-02",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgVnaLSupMCiWmiScxfMLx3.eval",
      "sourceId": "gVnaLSupMCiWmiScxfMLx3",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "granite-4.0-1b",
      "score": 0.008333333333333333,
      "bestScore": 0.008333333333333333,
      "standardError": 0.004700633956814717,
      "startedAt": "2026-08-30T22:46:30.000Z",
      "releaseDate": "2025-10-28",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPiFfhBoFbnaoqX5z2gVvra.eval",
      "sourceId": "PiFfhBoFbnaoqX5z2gVvra",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.5-35b-a3b_none",
      "score": 0.7,
      "bestScore": 0.7,
      "standardError": 0.06003996985174109,
      "startedAt": "2026-08-30T22:45:56.000Z",
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fh9aoDxyWEhybvARHpyuoco.eval",
      "sourceId": "h9aoDxyWEhybvARHpyuoco",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.5-35b-a3b",
      "score": 0.5444444444444444,
      "bestScore": 0.5444444444444444,
      "standardError": 0.061949030495634445,
      "startedAt": "2026-08-30T22:45:36.000Z",
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9WS6kHXT8XFtJWtE2wWLAo.eval",
      "sourceId": "9WS6kHXT8XFtJWtE2wWLAo",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Qwen3-32B_none",
      "score": 0.23055555555555557,
      "bestScore": 0.23055555555555557,
      "standardError": 0.05148605337836994,
      "startedAt": "2026-08-30T22:45:19.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCfmvbqzjpF6cCJQfzqdD4w.eval",
      "sourceId": "CfmvbqzjpF6cCJQfzqdD4w",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Qwen3-32B",
      "score": 0.6694444444444444,
      "bestScore": 0.6694444444444444,
      "standardError": 0.06040106323530329,
      "startedAt": "2026-08-30T22:44:49.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FBAi4jidcT3F6reBkSXDJKv.eval",
      "sourceId": "BAi4jidcT3F6reBkSXDJKv",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3-30b-a3b-thinking-2507",
      "score": 0.7027777777777777,
      "bestScore": 0.7027777777777777,
      "standardError": 0.060418480757624834,
      "startedAt": "2026-08-30T22:44:45.000Z",
      "releaseDate": "2025-07-30",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6rZFxNR3GjfhWey6DzEiYz.eval",
      "sourceId": "6rZFxNR3GjfhWey6DzEiYz",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3-30b-a3b-instruct-2507",
      "score": 0.6222222222222222,
      "bestScore": 0.6222222222222222,
      "standardError": 0.06509619058885623,
      "startedAt": "2026-08-30T22:44:11.000Z",
      "releaseDate": "2025-07-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FEo5cAQJK28CHpKDfCSDYG4.eval",
      "sourceId": "Eo5cAQJK28CHpKDfCSDYG4",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Qwen3-30B-A3B_none",
      "score": 0.25555555555555554,
      "bestScore": 0.25555555555555554,
      "standardError": 0.052095537290747576,
      "startedAt": "2026-08-30T22:43:50.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZzbWv2QGVfg4o2DTX45nqV.eval",
      "sourceId": "ZzbWv2QGVfg4o2DTX45nqV",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3-14b_none",
      "score": 0.25833333333333336,
      "bestScore": 0.25833333333333336,
      "standardError": 0.048232653761625936,
      "startedAt": "2026-08-30T22:42:57.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRMcys99fv45vvgTgakXXWn.eval",
      "sourceId": "RMcys99fv45vvgTgakXXWn",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen2.5-7b-instruct",
      "score": 0.025,
      "bestScore": 0.025,
      "standardError": 0.009401267913629434,
      "startedAt": "2026-08-30T22:42:37.000Z",
      "releaseDate": "2024-09-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FU8HHNLLLVHJ7sHCfwdiDBX.eval",
      "sourceId": "U8HHNLLLVHJ7sHCfwdiDBX",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "mistral-small-3.2-2506",
      "score": 0.30277777777777776,
      "bestScore": 0.30277777777777776,
      "standardError": 0.057774499712758075,
      "startedAt": "2026-08-30T22:41:17.000Z",
      "releaseDate": "2025-06-20",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FHyKzRUkV42yTmGrbYQvZxQ.eval",
      "sourceId": "HyKzRUkV42yTmGrbYQvZxQ",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "mistral-small-3.1-2503",
      "score": 0.03888888888888889,
      "bestScore": 0.03888888888888889,
      "standardError": 0.01529498571765449,
      "startedAt": "2026-08-30T22:41:08.000Z",
      "releaseDate": "2025-03-17",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FnNZHFNyHgfv3uzPAxspaPT.eval",
      "sourceId": "nNZHFNyHgfv3uzPAxspaPT",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "mistral-small-3-2501",
      "score": 0.06666666666666667,
      "bestScore": 0.06666666666666667,
      "standardError": 0.02682716849943299,
      "startedAt": "2026-08-30T22:40:34.000Z",
      "releaseDate": "2025-01-30",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLEBmfabfasGXFwRwhN9j35.eval",
      "sourceId": "LEBmfabfasGXFwRwhN9j35",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Mistral-7B-Instruct-v0.3",
      "score": 0.002777777777777778,
      "bestScore": 0.002777777777777778,
      "standardError": 0.002777777777777778,
      "startedAt": "2026-08-30T22:40:14.000Z",
      "releaseDate": "2024-05-27",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fj6Mo25oV9Y9jdGxCMHZfpU.eval",
      "sourceId": "j6Mo25oV9Y9jdGxCMHZfpU",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "magistral-small-2509",
      "score": 0.28055555555555556,
      "bestScore": 0.28055555555555556,
      "standardError": 0.0556407175551872,
      "startedAt": "2026-08-30T22:39:41.000Z",
      "releaseDate": "2025-09-18",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcSTYZ3FKp5Mbs3xopn68NK.eval",
      "sourceId": "cSTYZ3FKp5Mbs3xopn68NK",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Llama-3.2-1B-Instruct",
      "score": 0.005555555555555556,
      "bestScore": 0.005555555555555556,
      "standardError": 0.003883473863441843,
      "startedAt": "2026-08-30T22:39:12.000Z",
      "releaseDate": "2024-09-24",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F8qwT2CF9U5zV99PFirnVsp.eval",
      "sourceId": "8qwT2CF9U5zV99PFirnVsp",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Meta-Llama-3-8B-Instruct",
      "score": 0.019444444444444445,
      "bestScore": 0.019444444444444445,
      "standardError": 0.010477515876628662,
      "startedAt": "2026-08-30T22:38:53.000Z",
      "releaseDate": "2024-04-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCAcNPzYBEu39HFWPmuUEeP.eval",
      "sourceId": "CAcNPzYBEu39HFWPmuUEeP",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "glm-4.7-flash",
      "score": 0.5833333333333334,
      "bestScore": 0.5833333333333334,
      "standardError": 0.06564078093085213,
      "startedAt": "2026-08-30T22:37:43.000Z",
      "releaseDate": null,
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9ekki5tJhyjmA7ZMLoUxzo.eval",
      "sourceId": "9ekki5tJhyjmA7ZMLoUxzo",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-6-astra_max",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-08-30T14:57:43.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "L7kW8yxFKCjVXhxNnt3VTq",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "deepseek-llm-67b-chat",
      "score": 0.008333333333333333,
      "bestScore": 0.008333333333333333,
      "standardError": 0.004700633956814719,
      "startedAt": "2026-08-28T18:57:38.000Z",
      "releaseDate": "2023-11-29",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FY5rC5REnP83Q7mAoMtjiRq.eval",
      "sourceId": "Y5rC5REnP83Q7mAoMtjiRq",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "seed-oss-36b-instruct",
      "score": 0.675,
      "bestScore": 0.675,
      "standardError": 0.06324056203789373,
      "startedAt": "2026-08-28T18:40:22.000Z",
      "releaseDate": "2025-08-20",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FeJvV6RezECZwh3b4ekg4pm.eval",
      "sourceId": "eJvV6RezECZwh3b4ekg4pm",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.5-4b_none",
      "score": 0.5583333333333333,
      "bestScore": 0.5583333333333333,
      "standardError": 0.0616226296489243,
      "startedAt": "2026-08-28T18:33:35.000Z",
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSAZy8rk2HgyPjA8XZJFUmR.eval",
      "sourceId": "SAZy8rk2HgyPjA8XZJFUmR",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3-4b-instruct-2507",
      "score": 0.5222222222222223,
      "bestScore": 0.5222222222222223,
      "standardError": 0.06591537955214949,
      "startedAt": "2026-08-28T17:48:22.000Z",
      "releaseDate": "2025-08-06",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKH7KfZvFc9kZbqiJFdfYTc.eval",
      "sourceId": "KH7KfZvFc9kZbqiJFdfYTc",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Qwen3-30B-A3B",
      "score": 0.6277777777777778,
      "bestScore": 0.6277777777777778,
      "standardError": 0.062118645754659016,
      "startedAt": "2026-08-28T17:03:13.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXfjRkSyP4ub4gDsWcgS7KY.eval",
      "sourceId": "XfjRkSyP4ub4gDsWcgS7KY",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3-14b",
      "score": 0.6638888888888889,
      "bestScore": 0.6638888888888889,
      "standardError": 0.06190655406811223,
      "startedAt": "2026-08-28T16:54:14.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgRRQHECkYhz5UaYsRKDh2F.eval",
      "sourceId": "gRRQHECkYhz5UaYsRKDh2F",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3-1.7b_none",
      "score": 0.08055555555555556,
      "bestScore": 0.08055555555555556,
      "standardError": 0.0281353251405273,
      "startedAt": "2026-08-28T16:47:54.000Z",
      "releaseDate": "2025-04-29",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbbqokudairnKtbyjBjuWCx.eval",
      "sourceId": "bbqokudairnKtbyjBjuWCx",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemma-4-26b-a4b-it_minimal",
      "score": 0.8222222222222222,
      "bestScore": 0.8222222222222222,
      "standardError": 0.05763774795025092,
      "startedAt": "2026-08-28T12:55:56.000Z",
      "releaseDate": "2026-04-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmstwSH7noJfvhGcd6Lz5kd.eval",
      "sourceId": "mstwSH7noJfvhGcd6Lz5kd",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "granite-4.0-350m",
      "score": 0.002777777777777778,
      "bestScore": 0.002777777777777778,
      "standardError": 0.002777777777777778,
      "startedAt": "2026-08-28T12:37:51.000Z",
      "releaseDate": "2025-10-28",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FaQYiADNH3qVzZkWUmQn6nZ.eval",
      "sourceId": "aQYiADNH3qVzZkWUmQn6nZ",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "glm-4.7-flash_none",
      "score": 0.25,
      "bestScore": 0.25,
      "standardError": 0.049460472943941645,
      "startedAt": "2026-08-28T11:17:11.000Z",
      "releaseDate": null,
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfPE4Zajhq5WfccSxEYXXR7.eval",
      "sourceId": "fPE4Zajhq5WfccSxEYXXR7",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemma-3-4b-it",
      "score": 0.075,
      "bestScore": 0.075,
      "standardError": 0.02763853991962833,
      "startedAt": "2026-08-28T08:35:58.000Z",
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKAPGpH2CLMafAt8MnAQJT3.eval",
      "sourceId": "KAPGpH2CLMafAt8MnAQJT3",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemma-3-1b-it",
      "score": 0.011111111111111112,
      "bestScore": 0.011111111111111112,
      "standardError": 0.005362818328685681,
      "startedAt": "2026-08-28T08:28:18.000Z",
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F3CGChL5vNDM6Df5C8Z8owN.eval",
      "sourceId": "3CGChL5vNDM6Df5C8Z8owN",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemma-3-12b-it",
      "score": 0.16666666666666666,
      "bestScore": 0.16666666666666666,
      "standardError": 0.0424174778156649,
      "startedAt": "2026-08-28T08:22:55.000Z",
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCAyLyhDHZiMQBZDB3qd8tG.eval",
      "sourceId": "CAyLyhDHZiMQBZDB3qd8tG",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "DeepSeek-R1-Distill-Qwen-32B",
      "score": 0.5555555555555556,
      "bestScore": 0.5555555555555556,
      "standardError": 0.061999963795652153,
      "startedAt": "2026-08-28T08:15:43.000Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKbUNQf8nnwPa8ixg6AnEwB.eval",
      "sourceId": "KbUNQf8nnwPa8ixg6AnEwB",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "DeepSeek-R1-Distill-Qwen-14B",
      "score": 0.5055555555555555,
      "bestScore": 0.5055555555555555,
      "standardError": 0.05984981671790839,
      "startedAt": "2026-08-28T08:02:29.000Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FEceR3NeHzSxacdmAmfRxgT.eval",
      "sourceId": "EceR3NeHzSxacdmAmfRxgT",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "DeepSeek-R1-Distill-Qwen-1.5B",
      "score": 0.21388888888888888,
      "bestScore": 0.21388888888888888,
      "standardError": 0.05279771021800918,
      "startedAt": "2026-08-28T07:53:04.000Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGs234P3jzGiLm68KjyzYZL.eval",
      "sourceId": "Gs234P3jzGiLm68KjyzYZL",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "deepseek-r1-0528-qwen3-8b",
      "score": 0.4388888888888889,
      "bestScore": 0.4388888888888889,
      "standardError": 0.0677262181766212,
      "startedAt": "2026-08-28T07:49:02.000Z",
      "releaseDate": "2025-05-28",
      "organization": null,
      "country": null,
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLWbCXNGMXtjQYxyM2uoFMV.eval",
      "sourceId": "LWbCXNGMXtjQYxyM2uoFMV",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "QwQ-32B",
      "score": 0.5916666666666667,
      "bestScore": 0.5916666666666667,
      "standardError": 0.06348963965431165,
      "startedAt": "2026-08-28T07:20:55.000Z",
      "releaseDate": "2025-03-05",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2xMn6dbY9mw4pmkh8Tz9Gn.eval",
      "sourceId": "2xMn6dbY9mw4pmkh8Tz9Gn",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Llama-3.1-8B-Instruct",
      "score": 0.016666666666666666,
      "bestScore": 0.016666666666666666,
      "standardError": 0.009401267913629434,
      "startedAt": "2026-08-27T16:29:11.000Z",
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6YAzFq89fwWiJ3uJdptCMr.eval",
      "sourceId": "6YAzFq89fwWiJ3uJdptCMr",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.5-9B_none",
      "score": 0.6166666666666667,
      "bestScore": 0.6166666666666667,
      "standardError": 0.06003996985174109,
      "startedAt": "2026-08-27T14:31:08.000Z",
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYBMTDnW5LmCN6tBp2QYHzi.eval",
      "sourceId": "YBMTDnW5LmCN6tBp2QYHzi",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.5-9B",
      "score": 0.42777777777777776,
      "bestScore": 0.42777777777777776,
      "standardError": 0.06533819469304575,
      "startedAt": "2026-08-27T14:30:39.000Z",
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F5FVmQUSS4uRd23f6nNgeRe.eval",
      "sourceId": "5FVmQUSS4uRd23f6nNgeRe",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3-8b_none",
      "score": 0.2222222222222222,
      "bestScore": 0.2222222222222222,
      "standardError": 0.0452770806168054,
      "startedAt": "2026-08-27T14:22:01.000Z",
      "releaseDate": "2025-04-28",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FASLdzByzxxMxFeLc7dwTPi.eval",
      "sourceId": "ASLdzByzxxMxFeLc7dwTPi",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3-8b",
      "score": 0.5611111111111111,
      "bestScore": 0.5611111111111111,
      "standardError": 0.0645035213369854,
      "startedAt": "2026-08-27T14:21:53.000Z",
      "releaseDate": "2025-04-28",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgG2B6fLUDBy5jRraxL2pz8.eval",
      "sourceId": "gG2B6fLUDBy5jRraxL2pz8",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemma-3-27b-it",
      "score": 0.225,
      "bestScore": 0.225,
      "standardError": 0.052794388667238824,
      "startedAt": "2026-08-27T14:17:59.000Z",
      "releaseDate": "2025-03-12",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDhJ248BX3ZPBHXvkRG4Qsz.eval",
      "sourceId": "DhJ248BX3ZPBHXvkRG4Qsz",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-oss-20b_high",
      "score": 0.5083333333333333,
      "bestScore": 0.5083333333333333,
      "standardError": 0.06711496995394688,
      "startedAt": "2026-08-27T14:13:32.000Z",
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGMngscbSmqoPbGZWNv2Jvw.eval",
      "sourceId": "GMngscbSmqoPbGZWNv2Jvw",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-oss-20b_medium",
      "score": 0.6527777777777778,
      "bestScore": 0.6527777777777778,
      "standardError": 0.0593111883162336,
      "startedAt": "2026-08-27T14:12:40.000Z",
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDB8ja6pEZWGocQbdciyisL.eval",
      "sourceId": "DB8ja6pEZWGocQbdciyisL",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-oss-20b_low",
      "score": 0.4027777777777778,
      "bestScore": 0.4027777777777778,
      "standardError": 0.06317675132490254,
      "startedAt": "2026-08-27T14:12:25.000Z",
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcMwETYmC6b4yKupyJfEHv2.eval",
      "sourceId": "cMwETYmC6b4yKupyJfEHv2",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "glm-5.3-flash_max",
      "score": 0.9388888888888889,
      "bestScore": 0.9388888888888889,
      "standardError": 0.021622267380853318,
      "startedAt": "2026-08-26T22:26:55.000Z",
      "releaseDate": "2026-08-20",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "NCikvVWc5SEyMEbCqvVNb2",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "glm-5.3_max",
      "score": 0.9111111111111111,
      "bestScore": 0.9111111111111111,
      "standardError": 0.03454754040013868,
      "startedAt": "2026-08-24T19:44:18.000Z",
      "releaseDate": "2026-08-14",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "goZR5yGxZic9TZ23ezCL3w",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "deepseek-v4-pro-0813_max",
      "score": 0.9861111111111112,
      "bestScore": 0.9861111111111112,
      "standardError": 0.00990980650017605,
      "startedAt": "2026-08-18T21:03:23.000Z",
      "releaseDate": "2026-08-13",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "4HVvM9ZWKKPLQPcC8WYYMT",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.7-flash_high",
      "score": 0.9722222222222222,
      "bestScore": 0.9722222222222222,
      "standardError": 0.018586784946171872,
      "startedAt": "2026-08-14T18:29:12.000Z",
      "releaseDate": "2026-08-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "kQjrYCaa7yYyMXofeR23Dd",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "inkling-small_xhigh",
      "score": 0.9,
      "bestScore": 0.9,
      "standardError": 0.03964807305493795,
      "startedAt": "2026-08-14T14:19:47.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "53BDFebxiT9KGeP5FRt6bJ",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "grok-4.6_xhigh",
      "score": 0.9916666666666667,
      "bestScore": 0.9916666666666667,
      "standardError": 0.004700633956814717,
      "startedAt": "2026-08-14T13:47:42.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "7crLvnfQc987dC6mr3iGbf",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "grok-4.6_high",
      "score": 0.9777777777777777,
      "bestScore": 0.9777777777777777,
      "standardError": 0.00996275330534286,
      "startedAt": "2026-08-12T19:39:52.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "jtfdCURnwfZzak4BSiBKzx",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "nemotron-3-ultra",
      "score": 0.8666666666666667,
      "bestScore": 0.8666666666666667,
      "standardError": 0.051247074319053836,
      "startedAt": "2026-08-10T17:47:10.000Z",
      "releaseDate": "2026-06-04",
      "organization": "Nvidia",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWYTWr3Xk4qxymhMsm7vRQ2.eval",
      "sourceId": "WYTWr3Xk4qxymhMsm7vRQ2",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "MiniMax-M3_none",
      "score": 0.26666666666666666,
      "bestScore": 0.26666666666666666,
      "standardError": 0.06666666666666668,
      "startedAt": "2026-08-10T17:46:50.000Z",
      "releaseDate": "2026-06-01",
      "organization": "MiniMax",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F97t77hRrNae6Va8fBqBhPy.eval",
      "sourceId": "97t77hRrNae6Va8fBqBhPy",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "MiniMax-M3",
      "score": 0.7111111111111111,
      "bestScore": 0.7111111111111111,
      "standardError": 0.0683294324254051,
      "startedAt": "2026-08-10T17:46:09.000Z",
      "releaseDate": "2026-06-01",
      "organization": "MiniMax",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F79ZP6iwoBaF5t9a5VRqPnw.eval",
      "sourceId": "79ZP6iwoBaF5t9a5VRqPnw",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "glm-5.2_none",
      "score": 0.28888888888888886,
      "bestScore": 0.28888888888888886,
      "standardError": 0.06832943242540508,
      "startedAt": "2026-08-10T17:45:48.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJ8hYKbXiGhetQRwu5bUSp6.eval",
      "sourceId": "J8hYKbXiGhetQRwu5bUSp6",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "glm-5.2_low",
      "score": 0.7555555555555555,
      "bestScore": 0.7555555555555555,
      "standardError": 0.06478835438717,
      "startedAt": "2026-08-10T17:45:24.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9uKmMPn6eiRPsfcLKod7M7.eval",
      "sourceId": "9uKmMPn6eiRPsfcLKod7M7",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "glm-5.1",
      "score": 0.9333333333333333,
      "bestScore": 0.9333333333333333,
      "standardError": 0.03760507165451774,
      "startedAt": "2026-08-10T17:44:52.000Z",
      "releaseDate": "2026-04-07",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fb7Lw9ExxAhVY3hdExoh4MP.eval",
      "sourceId": "b7Lw9ExxAhVY3hdExoh4MP",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.4-mini-2026-03-17_xhigh",
      "score": 0.8888888888888888,
      "bestScore": 0.8888888888888888,
      "standardError": 0.047377936967913425,
      "startedAt": "2026-08-07T00:56:59.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQGMk6JNCYuUkFMF4zdVCsM.eval",
      "sourceId": "QGMk6JNCYuUkFMF4zdVCsM",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.6-sol_none",
      "score": 0.6888888888888889,
      "bestScore": 0.6888888888888889,
      "standardError": 0.0697920592732311,
      "startedAt": "2026-08-07T00:54:12.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fg7czyLHudHegzP9yRD939X.eval",
      "sourceId": "g7czyLHudHegzP9yRD939X",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.5_none",
      "score": 0.5777777777777777,
      "bestScore": 0.5777777777777777,
      "standardError": 0.07446027270295805,
      "startedAt": "2026-08-07T00:53:58.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQuqPmMUTbiEsUvoFYZx5Mk.eval",
      "sourceId": "QuqPmMUTbiEsUvoFYZx5Mk",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "kimi-k3_low",
      "score": 0.6888888888888889,
      "bestScore": 0.6888888888888889,
      "standardError": 0.06979205927323111,
      "startedAt": "2026-08-07T00:46:06.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfrszScNxpHXXjSHawCt9Zb.eval",
      "sourceId": "frszScNxpHXXjSHawCt9Zb",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "kimi-k3_high",
      "score": 0.9333333333333333,
      "bestScore": 0.9333333333333333,
      "standardError": 0.03760507165451775,
      "startedAt": "2026-08-07T00:46:05.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F4E5xGsuRMwMBhAfC4F2G4N.eval",
      "sourceId": "4E5xGsuRMwMBhAfC4F2G4N",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.6-terra_none",
      "score": 0.5333333333333333,
      "bestScore": 0.5333333333333333,
      "standardError": 0.07521014330903548,
      "startedAt": "2026-08-07T00:42:38.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcjUVqF2qQJ4qhFKBU5BQpW.eval",
      "sourceId": "cjUVqF2qQJ4qhFKBU5BQpW",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.4-mini-2026-03-17_none",
      "score": 0.26666666666666666,
      "bestScore": 0.26666666666666666,
      "standardError": 0.06666666666666668,
      "startedAt": "2026-08-07T00:42:28.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FR4NrJwtfdSQkGCWyKnrxgh.eval",
      "sourceId": "R4NrJwtfdSQkGCWyKnrxgh",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.6-max-preview",
      "score": 0.9111111111111111,
      "bestScore": 0.9111111111111111,
      "standardError": 0.04290254662948546,
      "startedAt": "2026-08-07T00:42:24.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FEfJpx5HztMmL5EQSwsyqMe.eval",
      "sourceId": "EfJpx5HztMmL5EQSwsyqMe",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.6-sol_low",
      "score": 0.9555555555555556,
      "bestScore": 0.9555555555555556,
      "standardError": 0.031067790907534736,
      "startedAt": "2026-08-07T00:42:21.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMifz96Jm442bn43Qnxj5rT.eval",
      "sourceId": "Mifz96Jm442bn43Qnxj5rT",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.5_low",
      "score": 0.8444444444444444,
      "bestScore": 0.8444444444444444,
      "standardError": 0.05463890236888291,
      "startedAt": "2026-08-07T00:39:05.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCdmtwZ8NR8YeTXWiLVTRDJ.eval",
      "sourceId": "CdmtwZ8NR8YeTXWiLVTRDJ",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.7-max",
      "score": 0.9555555555555556,
      "bestScore": 0.9555555555555556,
      "standardError": 0.031067790907534736,
      "startedAt": "2026-08-07T00:39:02.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FD3YDUZSfEUDsS4frLbEKDu.eval",
      "sourceId": "D3YDUZSfEUDsS4frLbEKDu",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o3-2025-04-16_medium",
      "score": 0.8444444444444444,
      "bestScore": 0.8444444444444444,
      "standardError": 0.05463890236888291,
      "startedAt": "2026-08-07T00:28:27.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F8GjRH3T4wVQMnBKX4fv4ub.eval",
      "sourceId": "8GjRH3T4wVQMnBKX4fv4ub",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.1-2025-11-13_none",
      "score": 0.37777777777777777,
      "bestScore": 0.37777777777777777,
      "standardError": 0.07309112127323451,
      "startedAt": "2026-08-07T00:24:39.000Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FH5NgPXUD57sbusTLp3yKQE.eval",
      "sourceId": "H5NgPXUD57sbusTLp3yKQE",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "kimi-k2.7-code",
      "score": 0.9555555555555556,
      "bestScore": 0.9555555555555556,
      "standardError": 0.031067790907534736,
      "startedAt": "2026-08-07T00:20:43.000Z",
      "releaseDate": "2026-06-12",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FENc5ZM5TpBYNKkTWN4xyQM.eval",
      "sourceId": "ENc5ZM5TpBYNKkTWN4xyQM",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.6-terra_low",
      "score": 0.8888888888888888,
      "bestScore": 0.8888888888888888,
      "standardError": 0.04737793696791343,
      "startedAt": "2026-08-07T00:20:21.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZ6S3nMC2tnkpKUixCPLiDj.eval",
      "sourceId": "Z6S3nMC2tnkpKUixCPLiDj",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.6-luna_none",
      "score": 0.4,
      "bestScore": 0.4,
      "standardError": 0.07385489458759964,
      "startedAt": "2026-08-07T00:20:15.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQYZ6wRuT7WMkaYPKc9dRPy.eval",
      "sourceId": "QYZ6wRuT7WMkaYPKc9dRPy",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.6-plus",
      "score": 0.9333333333333333,
      "bestScore": 0.9333333333333333,
      "standardError": 0.03760507165451774,
      "startedAt": "2026-08-07T00:16:48.000Z",
      "releaseDate": "2026-03-31",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjYiGAFfUQKJuRStPzU3PX3.eval",
      "sourceId": "jYiGAFfUQKJuRStPzU3PX3",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o4-mini-2025-04-16_medium",
      "score": 0.7333333333333333,
      "bestScore": 0.7333333333333333,
      "standardError": 0.06666666666666667,
      "startedAt": "2026-08-07T00:16:13.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTWJtwC5bzXa9HorTH8DNJy.eval",
      "sourceId": "TWJtwC5bzXa9HorTH8DNJy",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.6-27b",
      "score": 0.9111111111111111,
      "bestScore": 0.9111111111111111,
      "standardError": 0.04290254662948546,
      "startedAt": "2026-08-07T00:12:36.000Z",
      "releaseDate": "2026-04-22",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F452pB5w5ea6uy92e2q2NMv.eval",
      "sourceId": "452pB5w5ea6uy92e2q2NMv",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.6-27b_none",
      "score": 0.6666666666666666,
      "bestScore": 0.6666666666666666,
      "standardError": 0.07106690545187015,
      "startedAt": "2026-08-07T00:12:23.000Z",
      "releaseDate": "2026-04-22",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fo6VtA8qruLpoShhPy2JdLg.eval",
      "sourceId": "o6VtA8qruLpoShhPy2JdLg",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.5-397b-a17b",
      "score": 0.8888888888888888,
      "bestScore": 0.8888888888888888,
      "standardError": 0.04737793696791343,
      "startedAt": "2026-08-07T00:12:18.000Z",
      "releaseDate": "2026-02-13",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJYuFdyZZSiMvmSZNzMTywb.eval",
      "sourceId": "JYuFdyZZSiMvmSZNzMTywb",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.5-397b-a17b_none",
      "score": 0.8222222222222222,
      "bestScore": 0.8222222222222222,
      "standardError": 0.057637747950250916,
      "startedAt": "2026-08-07T00:12:04.000Z",
      "releaseDate": "2026-02-13",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fi2ZUgMREKiFAnE2M54w3Py.eval",
      "sourceId": "i2ZUgMREKiFAnE2M54w3Py",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.5-plus",
      "score": 0.8666666666666667,
      "bestScore": 0.8666666666666667,
      "standardError": 0.051247074319053836,
      "startedAt": "2026-08-07T00:11:58.000Z",
      "releaseDate": "2026-02-16",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcFbntXuxUtfreULvt7vEt7.eval",
      "sourceId": "cFbntXuxUtfreULvt7vEt7",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.6-flash_minimal",
      "score": 0.8,
      "bestScore": 0.8,
      "standardError": 0.06030226891555273,
      "startedAt": "2026-08-07T00:07:32.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcF97JqHbFxxp7ePjYVK9HV.eval",
      "sourceId": "cF97JqHbFxxp7ePjYVK9HV",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.6-flash_low",
      "score": 0.8222222222222222,
      "bestScore": 0.8222222222222222,
      "standardError": 0.057637747950250916,
      "startedAt": "2026-08-07T00:07:20.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6iEkMa88YjLDgcde4bLm8B.eval",
      "sourceId": "6iEkMa88YjLDgcde4bLm8B",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.6-35b-a3b",
      "score": 0.8666666666666667,
      "bestScore": 0.8666666666666667,
      "standardError": 0.051247074319053836,
      "startedAt": "2026-08-07T00:03:16.000Z",
      "releaseDate": "2026-04-14",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F3n5KDcBaKqJh3WTg7qU2rs.eval",
      "sourceId": "3n5KDcBaKqJh3WTg7qU2rs",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.6-35b-a3b_none",
      "score": 0.6888888888888889,
      "bestScore": 0.6888888888888889,
      "standardError": 0.06979205927323111,
      "startedAt": "2026-08-07T00:03:06.000Z",
      "releaseDate": "2026-04-14",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7GiYWM7UagUsHpNsdymd4z.eval",
      "sourceId": "7GiYWM7UagUsHpNsdymd4z",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.7-plus",
      "score": 0.9333333333333333,
      "bestScore": 0.9333333333333333,
      "standardError": 0.03760507165451775,
      "startedAt": "2026-08-07T00:02:57.000Z",
      "releaseDate": "2026-06-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJvdw6GAXiS6T79bkpbk6Ra.eval",
      "sourceId": "Jvdw6GAXiS6T79bkpbk6Ra",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.7-plus_none",
      "score": 0.8,
      "bestScore": 0.8,
      "standardError": 0.06030226891555273,
      "startedAt": "2026-08-07T00:02:56.000Z",
      "releaseDate": "2026-06-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fo5th8Rb9WiaDbv4oEMDoFa.eval",
      "sourceId": "o5th8Rb9WiaDbv4oEMDoFa",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.7-flash",
      "score": 0.8666666666666667,
      "bestScore": 0.8666666666666667,
      "standardError": 0.051247074319053836,
      "startedAt": "2026-08-07T00:02:46.000Z",
      "releaseDate": "2026-07-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTjoS7sNuqVTc4LVyGRo4wa.eval",
      "sourceId": "TjoS7sNuqVTc4LVyGRo4wa",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.7-flash_none",
      "score": 0.7777777777777778,
      "bestScore": 0.7777777777777778,
      "standardError": 0.06267511942419625,
      "startedAt": "2026-08-07T00:02:33.000Z",
      "releaseDate": "2026-07-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSZDgstFQJAGeGQwMXgNx6w.eval",
      "sourceId": "SZDgstFQJAGeGQwMXgNx6w",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.6-luna_low",
      "score": 0.6666666666666666,
      "bestScore": 0.6666666666666666,
      "standardError": 0.07106690545187014,
      "startedAt": "2026-08-07T00:02:15.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZgU8BagFmaj7BE94y3JDZA.eval",
      "sourceId": "ZgU8BagFmaj7BE94y3JDZA",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.6-flash",
      "score": 0.8444444444444444,
      "bestScore": 0.8444444444444444,
      "standardError": 0.05463890236888291,
      "startedAt": "2026-08-07T00:02:08.000Z",
      "releaseDate": "2026-04-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9zx2EXPTvPmxsL9qFCjP3x.eval",
      "sourceId": "9zx2EXPTvPmxsL9qFCjP3x",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.4-nano-2026-03-17_none",
      "score": 0.4666666666666667,
      "bestScore": 0.4666666666666667,
      "standardError": 0.07521014330903548,
      "startedAt": "2026-08-07T00:01:03.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FeUR9o3bvU2Yu3KY5w5uk4j.eval",
      "sourceId": "eUR9o3bvU2Yu3KY5w5uk4j",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-3.5-turbo-0125",
      "score": 0.022222222222222223,
      "bestScore": 0.022222222222222223,
      "standardError": 0.022222222222222223,
      "startedAt": "2026-08-07T00:00:50.000Z",
      "releaseDate": "2024-01-25",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7ZVhkukwaRXU2Sx2ttibPF.eval",
      "sourceId": "7ZVhkukwaRXU2Sx2ttibPF",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.5-flash",
      "score": 0.8444444444444444,
      "bestScore": 0.8444444444444444,
      "standardError": 0.05463890236888291,
      "startedAt": "2026-08-07T00:00:21.000Z",
      "releaseDate": "2026-02-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FawDKyKbPXRzBVZzfdQucef.eval",
      "sourceId": "awDKyKbPXRzBVZzfdQucef",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5-mini-2025-08-07_minimal",
      "score": 0.5555555555555556,
      "bestScore": 0.5555555555555556,
      "standardError": 0.07491109582924914,
      "startedAt": "2026-08-07T00:00:04.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7VLdaSiAanBwM9TJUB5BhU.eval",
      "sourceId": "7VLdaSiAanBwM9TJUB5BhU",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.4-nano-2026-03-17_low",
      "score": 0.6888888888888889,
      "bestScore": 0.6888888888888889,
      "standardError": 0.06979205927323111,
      "startedAt": "2026-08-07T00:00:00.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbxvsuZ4Xpi3gZ4rsA8T7kV.eval",
      "sourceId": "bxvsuZ4Xpi3gZ4rsA8T7kV",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemma-4-31b-it_minimal",
      "score": 0.7333333333333333,
      "bestScore": 0.7333333333333333,
      "standardError": 0.06666666666666667,
      "startedAt": "2026-08-06T23:59:45.000Z",
      "releaseDate": "2026-04-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjFJucVunfLgoQqoFvPVQWP.eval",
      "sourceId": "jFJucVunfLgoQqoFvPVQWP",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.1-pro-preview_high",
      "score": 0.9555555555555556,
      "bestScore": 0.9555555555555556,
      "standardError": 0.031067790907534736,
      "startedAt": "2026-08-06T23:59:11.000Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fnhim2QCRgpir5cZdqrmUz3.eval",
      "sourceId": "nhim2QCRgpir5cZdqrmUz3",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "deepseek-v4-pro_none",
      "score": 0.4666666666666667,
      "bestScore": 0.4666666666666667,
      "standardError": 0.07521014330903548,
      "startedAt": "2026-08-06T23:59:05.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7bwsMdahVHrkaiN5x49Dk7.eval",
      "sourceId": "7bwsMdahVHrkaiN5x49Dk7",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "deepseek-v4-pro_high",
      "score": 0.9555555555555556,
      "bestScore": 0.9555555555555556,
      "standardError": 0.031067790907534743,
      "startedAt": "2026-08-06T23:58:58.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGLZkYFDFDENRiHyL68Nq43.eval",
      "sourceId": "GLZkYFDFDENRiHyL68Nq43",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-6_max",
      "score": 0.9111111111111111,
      "bestScore": 0.9111111111111111,
      "standardError": 0.04290254662948546,
      "startedAt": "2026-08-06T23:58:17.000Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcoxYk5vK5cBhQ7iZohMtA5.eval",
      "sourceId": "coxYk5vK5cBhQ7iZohMtA5",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.5-flash-lite_low",
      "score": 0.6,
      "bestScore": 0.6,
      "standardError": 0.07385489458759963,
      "startedAt": "2026-08-06T22:36:02.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FB3Kno3buYWRYbZxU33STiX.eval",
      "sourceId": "B3Kno3buYWRYbZxU33STiX",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.5-flash-lite_minimal",
      "score": 0.5111111111111111,
      "bestScore": 0.5111111111111111,
      "standardError": 0.07535922203472521,
      "startedAt": "2026-08-06T22:35:57.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fivr2LfmdZoCMfLME7jh2L8.eval",
      "sourceId": "ivr2LfmdZoCMfLME7jh2L8",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.5-flash-lite_high",
      "score": 0.7111111111111111,
      "bestScore": 0.7111111111111111,
      "standardError": 0.0683294324254051,
      "startedAt": "2026-08-06T22:35:36.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fmqsc4vFuKtAHXxbYyX9WwK.eval",
      "sourceId": "mqsc4vFuKtAHXxbYyX9WwK",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.5-flash_low",
      "score": 0.8888888888888888,
      "bestScore": 0.8888888888888888,
      "standardError": 0.047377936967913425,
      "startedAt": "2026-08-06T22:34:49.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fig7gQ23WRms9SQLLNPtbTi.eval",
      "sourceId": "ig7gQ23WRms9SQLLNPtbTi",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.1-flash-lite_minimal",
      "score": 0.37777777777777777,
      "bestScore": 0.37777777777777777,
      "standardError": 0.07309112127323451,
      "startedAt": "2026-08-06T22:30:17.000Z",
      "releaseDate": "2026-03-03",
      "organization": "Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FEoSFpgFhJB7iDPLcMNuLH4.eval",
      "sourceId": "EoSFpgFhJB7iDPLcMNuLH4",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.1-flash-lite_low",
      "score": 0.4444444444444444,
      "bestScore": 0.4444444444444444,
      "standardError": 0.07491109582924914,
      "startedAt": "2026-08-06T22:30:07.000Z",
      "releaseDate": "2026-03-03",
      "organization": "Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCFBCC2QrkCAFBAvFj5pqPF.eval",
      "sourceId": "CFBCC2QrkCAFBAvFj5pqPF",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.1-flash-lite_high",
      "score": 0.8,
      "bestScore": 0.8,
      "standardError": 0.06030226891555273,
      "startedAt": "2026-08-06T22:29:57.000Z",
      "releaseDate": "2026-03-03",
      "organization": "Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FNMdaAXyqqPRmWfuxZUfHFF.eval",
      "sourceId": "NMdaAXyqqPRmWfuxZUfHFF",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3-flash-preview_high",
      "score": 0.9555555555555556,
      "bestScore": 0.9555555555555556,
      "standardError": 0.031067790907534743,
      "startedAt": "2026-08-06T22:25:57.000Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fa4uE6KgMhGSUjcnemRnbkA.eval",
      "sourceId": "a4uE6KgMhGSUjcnemRnbkA",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-5_max",
      "score": 0.8,
      "bestScore": 0.8,
      "standardError": 0.06030226891555273,
      "startedAt": "2026-08-06T22:21:48.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXRutLYUVUePe69APTSTw8S.eval",
      "sourceId": "XRutLYUVUePe69APTSTw8S",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-5_low",
      "score": 0.9333333333333333,
      "bestScore": 0.9333333333333333,
      "standardError": 0.03760507165451774,
      "startedAt": "2026-08-06T22:21:20.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZ5FAHm8aqmk8NUUm7tojng.eval",
      "sourceId": "Z5FAHm8aqmk8NUUm7tojng",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-6_max",
      "score": 0.7111111111111111,
      "bestScore": 0.7111111111111111,
      "standardError": 0.06832943242540508,
      "startedAt": "2026-08-06T22:21:16.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXnQVH5qSfxVjpzoPwUEmnh.eval",
      "sourceId": "XnQVH5qSfxVjpzoPwUEmnh",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-5",
      "score": 0.9777777777777777,
      "bestScore": 0.9777777777777777,
      "standardError": 0.022222222222222223,
      "startedAt": "2026-08-06T22:21:15.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgsBidnvEcmCfLnn5sxePtk.eval",
      "sourceId": "gsBidnvEcmCfLnn5sxePtk",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-8_none",
      "score": 0.8444444444444444,
      "bestScore": 0.8444444444444444,
      "standardError": 0.05463890236888291,
      "startedAt": "2026-08-06T22:21:06.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FadGE8rXdNuUFs96ioppbc5.eval",
      "sourceId": "adGE8rXdNuUFs96ioppbc5",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-8_low",
      "score": 0.9777777777777777,
      "bestScore": 0.9777777777777777,
      "standardError": 0.022222222222222223,
      "startedAt": "2026-08-06T22:20:24.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMEhV5sh6nbM8cKJGQ6HzcS.eval",
      "sourceId": "MEhV5sh6nbM8cKJGQ6HzcS",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-7_max",
      "score": 0.8666666666666667,
      "bestScore": 0.8666666666666667,
      "standardError": 0.051247074319053836,
      "startedAt": "2026-08-06T22:19:46.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FT4y4iGHMqA8tXCdBKr5GjS.eval",
      "sourceId": "T4y4iGHMqA8tXCdBKr5GjS",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-fable-5_low",
      "score": 0.9777777777777777,
      "bestScore": 0.9777777777777777,
      "standardError": 0.022222222222222223,
      "startedAt": "2026-08-06T22:18:24.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6E989Xxeuw925T7cCV3XGB.eval",
      "sourceId": "6E989Xxeuw925T7cCV3XGB",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-fable-5_high",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-08-06T22:18:19.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhwrnvtEbDAocFtjcsHqsQh.eval",
      "sourceId": "hwrnvtEbDAocFtjcsHqsQh",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "magistral-small-2506",
      "score": 0.3,
      "bestScore": 0.3,
      "standardError": 0.05924610544247621,
      "startedAt": "2026-08-06T22:17:05.000Z",
      "releaseDate": "2025-06-10",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhmZPoiHBpqufv9r7ujiK8V.eval",
      "sourceId": "hmZPoiHBpqufv9r7ujiK8V",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Inkling_xhigh",
      "score": 0.8888888888888888,
      "bestScore": 0.8888888888888888,
      "standardError": 0.039171949640122654,
      "startedAt": "2026-08-05T18:55:39.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "j2fELMf364qfh2cpaVDL6W",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3.8-max_xhigh",
      "score": 0.9944444444444445,
      "bestScore": 0.9944444444444445,
      "standardError": 0.0038834738634418433,
      "startedAt": "2026-08-04T16:28:46.000Z",
      "releaseDate": "2026-08-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "aGRKHaq9uSawfYVpC565zU",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.5-instant",
      "score": 0.6805555555555556,
      "bestScore": 0.6805555555555556,
      "standardError": 0.062127114361967634,
      "startedAt": "2026-08-02T02:46:37.000Z",
      "releaseDate": "2026-05-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hGzLQWbxPdgQiiMbaDqF8k",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "deepseek-v4-flash-0731_max",
      "score": 0.9444444444444444,
      "bestScore": 0.9444444444444444,
      "standardError": 0.020854366486931897,
      "startedAt": "2026-08-02T00:22:31.000Z",
      "releaseDate": "2026-07-31",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "FQ3Suh2vKacz7UNqJybTWk",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.6-flash_high",
      "score": 0.9416666666666667,
      "bestScore": 0.9416666666666667,
      "standardError": 0.030926354359015452,
      "startedAt": "2026-08-02T00:21:27.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ajQFZGMjvyMTH2ULDZhwtz",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-5_max",
      "score": 0.9888888888888889,
      "bestScore": 0.9888888888888889,
      "standardError": 0.011111111111111112,
      "startedAt": "2026-07-24T18:40:39.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "KxvHAT9SV2dTDn7jBnTk6f",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o3-mini-2025-01-31_low",
      "score": 0.4444444444444444,
      "bestScore": 0.4444444444444444,
      "standardError": 0.07491109582924914,
      "startedAt": "2026-07-20T18:28:54.000Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDvMvnsvXgBB38PEnLGbTPH.eval",
      "sourceId": "DvMvnsvXgBB38PEnLGbTPH",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5-2025-08-07_minimal",
      "score": 0.4666666666666667,
      "bestScore": 0.4666666666666667,
      "standardError": 0.07521014330903548,
      "startedAt": "2026-07-20T18:28:50.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRMsr7t8ZwJmox4m863sJGm.eval",
      "sourceId": "RMsr7t8ZwJmox4m863sJGm",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "kimi-k3_max",
      "score": 0.9722222222222222,
      "bestScore": 0.9722222222222222,
      "standardError": 0.011158359139033182,
      "startedAt": "2026-07-16T21:59:40.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "A33om9MTYMC7xYeLv9X7Ha",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "deepseek-chat",
      "score": 0.4888888888888889,
      "bestScore": 0.4888888888888889,
      "standardError": 0.0753592220347252,
      "startedAt": "2026-07-16T00:10:43.000Z",
      "releaseDate": "2025-12-01",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDzXSCtsyMBfoBAhBDMxhsQ.eval",
      "sourceId": "DzXSCtsyMBfoBAhBDMxhsQ",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o1-2024-12-17_low",
      "score": 0.5333333333333333,
      "bestScore": 0.5333333333333333,
      "standardError": 0.07521014330903548,
      "startedAt": "2026-07-15T06:14:17.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJsAVnfoWmxCaChyiNU9VFi.eval",
      "sourceId": "JsAVnfoWmxCaChyiNU9VFi",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o1-2024-12-17_high",
      "score": 0.7333333333333333,
      "bestScore": 0.7333333333333333,
      "standardError": 0.06666666666666667,
      "startedAt": "2026-07-15T06:14:12.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FAEQDjKHZzrVUubU42ZUZwA.eval",
      "sourceId": "AEQDjKHZzrVUubU42ZUZwA",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o3-2025-04-16_low",
      "score": 0.6,
      "bestScore": 0.6,
      "standardError": 0.07385489458759963,
      "startedAt": "2026-07-15T03:15:08.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F8ygEN7GrcTGeBLbVWT9JQ8.eval",
      "sourceId": "8ygEN7GrcTGeBLbVWT9JQ8",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.4-2026-03-05_medium",
      "score": 0.9555555555555556,
      "bestScore": 0.9555555555555556,
      "standardError": 0.031067790907534736,
      "startedAt": "2026-07-15T03:14:58.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FeUJNMJSdLnJWNgRY8D2rYw.eval",
      "sourceId": "eUJNMJSdLnJWNgRY8D2rYw",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.4-2026-03-05_none",
      "score": 0.5777777777777777,
      "bestScore": 0.5777777777777777,
      "standardError": 0.07446027270295805,
      "startedAt": "2026-07-15T03:14:58.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Ffjxs2EWts9GjpuBtFD5BN3.eval",
      "sourceId": "fjxs2EWts9GjpuBtFD5BN3",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.4-2026-03-05_high",
      "score": 0.9777777777777777,
      "bestScore": 0.9777777777777777,
      "standardError": 0.022222222222222223,
      "startedAt": "2026-07-15T03:14:50.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDpPahiEawokzgidE3cNZbS.eval",
      "sourceId": "DpPahiEawokzgidE3cNZbS",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.4-2026-03-05_low",
      "score": 0.8444444444444444,
      "bestScore": 0.8444444444444444,
      "standardError": 0.05463890236888291,
      "startedAt": "2026-07-15T03:14:49.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbezRyWjXxqHnQeiL4MkSfZ.eval",
      "sourceId": "bezRyWjXxqHnQeiL4MkSfZ",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.5-flash_minimal",
      "score": 0.8,
      "bestScore": 0.8,
      "standardError": 0.06030226891555273,
      "startedAt": "2026-07-15T03:14:20.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJuM7FArkmhLqqKBqudySsC.eval",
      "sourceId": "JuM7FArkmhLqqKBqudySsC",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5-nano-2025-08-07_low",
      "score": 0.4666666666666667,
      "bestScore": 0.4666666666666667,
      "standardError": 0.07521014330903548,
      "startedAt": "2026-07-13T17:28:21.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F298stsbTPpZp7toZsS5cp8.eval",
      "sourceId": "298stsbTPpZp7toZsS5cp8",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-6_high",
      "score": 0.7555555555555555,
      "bestScore": 0.7555555555555555,
      "standardError": 0.06478835438717,
      "startedAt": "2026-07-13T17:28:15.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FoXZ7ZVZtADrwFPKAe6zTvL.eval",
      "sourceId": "oXZ7ZVZtADrwFPKAe6zTvL",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5-nano-2025-08-07_minimal",
      "score": 0.35555555555555557,
      "bestScore": 0.35555555555555557,
      "standardError": 0.0721639236343101,
      "startedAt": "2026-07-13T17:28:09.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FkJHhuWKGEEunTq9FxwKx9E.eval",
      "sourceId": "kJHhuWKGEEunTq9FxwKx9E",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.2-2025-12-11_none",
      "score": 0.6222222222222222,
      "bestScore": 0.6222222222222222,
      "standardError": 0.07309112127323451,
      "startedAt": "2026-07-13T17:28:07.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FUBKiF8QVUmHwqgx4rF7CDA.eval",
      "sourceId": "UBKiF8QVUmHwqgx4rF7CDA",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-6_medium",
      "score": 0.8222222222222222,
      "bestScore": 0.8222222222222222,
      "standardError": 0.057637747950250916,
      "startedAt": "2026-07-13T17:28:00.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FkDNNbUYXhrAH5uDBGykFaf.eval",
      "sourceId": "kDNNbUYXhrAH5uDBGykFaf",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o4-mini-2025-04-16_low",
      "score": 0.5777777777777777,
      "bestScore": 0.5777777777777777,
      "standardError": 0.07446027270295805,
      "startedAt": "2026-07-13T16:50:13.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWNYdVF9tnK7Nf6We9NEhJS.eval",
      "sourceId": "WNYdVF9tnK7Nf6We9NEhJS",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "grok-4.20-0309-reasoning",
      "score": 0.9222222222222223,
      "bestScore": 0.9222222222222223,
      "standardError": 0.032829739672413996,
      "startedAt": "2026-07-13T16:45:48.000Z",
      "releaseDate": "2026-02-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mM6vdJpgjLY9prnetXzZJS",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.6-terra_max",
      "score": 0.9972222222222222,
      "bestScore": 0.9972222222222222,
      "standardError": 0.002777777777777778,
      "startedAt": "2026-07-09T17:07:54.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "jUmmCNezoW7jWiK72SVCaP",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.6-luna_max",
      "score": 0.9833333333333333,
      "bestScore": 0.9833333333333333,
      "standardError": 0.011650421590325524,
      "startedAt": "2026-07-09T13:36:49.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "AhtQ5ggs4pTfLDqHFjueGy",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.6-sol_max",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-07-09T02:07:13.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "7RjsANKwwsePSstcZeSMhX",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "grok-4.5_high",
      "score": 0.9777777777777777,
      "bestScore": 0.9777777777777777,
      "standardError": 0.012743145865697103,
      "startedAt": "2026-07-08T23:54:22.000Z",
      "releaseDate": "2026-07-08",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "f8Pac6hqib4t7x2s4HCRoT",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-5_xhigh",
      "score": 0.9472222222222222,
      "bestScore": 0.9472222222222222,
      "standardError": 0.022644349282618412,
      "startedAt": "2026-07-01T16:06:23.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "T2wFfuSkUdyjpyqurxNmNK",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "glm-5.2_max",
      "score": 0.8638888888888889,
      "bestScore": 0.8638888888888889,
      "standardError": 0.04219779330510574,
      "startedAt": "2026-06-25T00:15:08.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "XjFqou6MyT2XLpkUsCrgxM",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "grok-4.3_high",
      "score": 0.9333333333333333,
      "bestScore": 0.9333333333333333,
      "standardError": 0.03041173683586219,
      "startedAt": "2026-06-17T17:51:54.000Z",
      "releaseDate": "2026-04-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "fzoUFmy77NpEDtXyA6w6ap",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "deepseek-v4-pro_max",
      "score": 0.9666666666666667,
      "bestScore": 0.9666666666666667,
      "standardError": 0.02002208376727653,
      "startedAt": "2026-06-17T00:34:52.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "HjTDENSpy9sancDzr7pccY",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-fable-5_max",
      "score": 0.9972222222222222,
      "bestScore": 0.9972222222222222,
      "standardError": 0.002777777777777778,
      "startedAt": "2026-06-10T20:03:17.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "eiLEpRUvK48wUKDh4sYkzV",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-8_max",
      "score": 0.9833333333333333,
      "bestScore": 0.9833333333333333,
      "standardError": 0.014101901870444154,
      "startedAt": "2026-06-07T23:44:54.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "37pJqPpXTE6ia9ikmmGPJ4",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.5-flash_high",
      "score": 0.9555555555555556,
      "bestScore": 0.9555555555555556,
      "standardError": 0.026696111689702044,
      "startedAt": "2026-05-25T13:10:18.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FeYmgjYJfM7Utm46nfoRfZ6.eval",
      "sourceId": "eYmgjYJfM7Utm46nfoRfZ6",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "kimi-k2.6",
      "score": 0.9611111111111111,
      "bestScore": 0.9611111111111111,
      "standardError": 0.023777636174410793,
      "startedAt": "2026-05-02T09:08:49.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FBTeRdhgt6TFBj6QkGX2j6P.eval",
      "sourceId": "BTeRdhgt6TFBj6QkGX2j6P",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.5-pro-pre-release_xhigh",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-04-24T23:38:07.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYxbseUDNgFUjokfYXui62r.eval",
      "sourceId": "YxbseUDNgFUjokfYXui62r",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.5-pre-release_xhigh",
      "score": 1,
      "bestScore": 1,
      "standardError": 0,
      "startedAt": "2026-04-24T14:06:46.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdP6L7ViZSdcTiWxooLCTUh.eval",
      "sourceId": "dP6L7ViZSdcTiWxooLCTUh",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-7_xhigh",
      "score": 0.978,
      "bestScore": 0.978,
      "standardError": 0.022,
      "startedAt": "2026-04-17T18:26:45.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjyCZnpXG2bh9G43X8pT75m.eval",
      "sourceId": "jyCZnpXG2bh9G43X8pT75m",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.4-mini-2026-03-17_high",
      "score": 0.8722222222222222,
      "bestScore": 0.8722222222222222,
      "standardError": 0.042508334166668306,
      "startedAt": "2026-04-15T14:17:38.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZ5Sp2AKYzTBNdMEyTfTRuN.eval",
      "sourceId": "Z5Sp2AKYzTBNdMEyTfTRuN",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.4-nano-2026-03-17_high",
      "score": 0.8778,
      "bestScore": 0.8778,
      "standardError": 0.0394,
      "startedAt": "2026-04-14T15:08:33.525Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjVv9m8DjkdpZaLPdUzAnG9.eval",
      "sourceId": "jVv9m8DjkdpZaLPdUzAnG9",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "muse-spark",
      "score": 0.889,
      "bestScore": 0.889,
      "standardError": 0.047,
      "startedAt": "2026-04-08T16:30:04.184Z",
      "releaseDate": "2026-04-08",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "muse-spark-pre-release-mock-aime",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.4-2026-03-05_xhigh",
      "score": 0.953,
      "bestScore": 0.953,
      "standardError": 0.032,
      "startedAt": "2026-03-06T03:24:03.099Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FHVsPchr34pWYHNuv5sKYyK.eval",
      "sourceId": "HVsPchr34pWYHNuv5sKYyK",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-6_32K",
      "score": 0.857936507936508,
      "bestScore": 0.857936507936508,
      "standardError": 0.04539927553202047,
      "startedAt": "2026-02-20T16:18:28.678Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9LWdzVN5w4ihNvncsaCWMu.eval",
      "sourceId": "9LWdzVN5w4ihNvncsaCWMu",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3.1-pro-preview",
      "score": 0.956,
      "bestScore": 0.956,
      "standardError": 0.031,
      "startedAt": "2026-02-20T04:39:36.998Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbXgZZgzRE3wXoSRUgrbsVf.eval",
      "sourceId": "bXgZZgzRE3wXoSRUgrbsVf",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "glm-5",
      "score": 0.8,
      "bestScore": 0.8,
      "standardError": 0.06,
      "startedAt": "2026-02-12T16:08:22.578Z",
      "releaseDate": "2026-02-11",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fccbx7KQY9bTD563vzBXwSg.eval",
      "sourceId": "ccbx7KQY9bTD563vzBXwSg",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-6_64K",
      "score": 0.9444444444444444,
      "bestScore": 0.9444444444444444,
      "standardError": 0.028247294597471075,
      "startedAt": "2026-02-06T04:32:30.354Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTUPC8VdX4jApFa4MGfNP4t.eval",
      "sourceId": "TUPC8VdX4jApFa4MGfNP4t",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-6_32K",
      "score": 0.9305555555555556,
      "bestScore": 0.9305555555555556,
      "standardError": 0.03337013961212755,
      "startedAt": "2026-02-06T04:31:07.514Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FkZ3VgrQJxfKQ5msjpmSaAz.eval",
      "sourceId": "kZ3VgrQJxfKQ5msjpmSaAz",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "fireworks/kimi-k2p5",
      "score": 0.922,
      "bestScore": 0.922,
      "standardError": 0.039,
      "startedAt": "2026-02-02T10:53:14.986Z",
      "releaseDate": "2026-01-27",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmbfZvrSNoXSJqCyD4VKqLe.eval",
      "sourceId": "mbfZvrSNoXSJqCyD4VKqLe",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "glm-4.7",
      "score": 0.8333333333333334,
      "bestScore": 0.8333333333333334,
      "standardError": 0.04767312946227961,
      "startedAt": "2026-01-29T16:46:50.435Z",
      "releaseDate": "2025-12-22",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGYzmrzMfeSjNjm2J6L9PeT.eval",
      "sourceId": "GYzmrzMfeSjNjm2J6L9PeT",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3-flash-preview",
      "score": 0.9277777777777778,
      "bestScore": 0.9277777777777778,
      "standardError": 0.035196259111492545,
      "startedAt": "2025-12-17T22:40:30.896Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FH3E7EjjXfGCPnm9ooLdmAo.eval",
      "sourceId": "H3E7EjjXfGCPnm9ooLdmAo",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "deepseek-reasoner",
      "score": 0.878174603174603,
      "bestScore": 0.878174603174603,
      "standardError": 0.041306766974863386,
      "startedAt": "2025-12-16T14:32:17.922Z",
      "releaseDate": "2025-12-01",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F8szQqvTsGG8gJ2TiFReRn9.eval",
      "sourceId": "8szQqvTsGG8gJ2TiFReRn9",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.2-2025-12-11_xhigh",
      "score": 0.961,
      "bestScore": 0.961,
      "standardError": 0.027,
      "startedAt": "2025-12-13T15:56:00.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FR393BZjXtkEYju29gs5R8B.eval",
      "sourceId": "R393BZjXtkEYju29gs5R8B",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "openai/gpt-oss-120b_high",
      "score": 0.8888888888888888,
      "bestScore": 0.8888888888888888,
      "standardError": 0.04463343655613273,
      "startedAt": "2025-12-11T22:37:41.702Z",
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F3QjP5HPi8tKUzHtNJktfv4.eval",
      "sourceId": "3QjP5HPi8tKUzHtNJktfv4",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.2-2025-12-11_medium",
      "score": 0.9388888888888889,
      "bestScore": 0.9388888888888889,
      "standardError": 0.030675842208062715,
      "startedAt": "2025-12-11T20:49:22.447Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZ2ZwwqLDDrBrEYhtKtFLCQ.eval",
      "sourceId": "Z2ZwwqLDDrBrEYhtKtFLCQ",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.2-2025-12-11_high",
      "score": 0.9611111111111111,
      "bestScore": 0.9611111111111111,
      "standardError": 0.025997226390952678,
      "startedAt": "2025-12-11T20:49:15.912Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLiUMZyNSM8q93KDubWgYCh.eval",
      "sourceId": "LiUMZyNSM8q93KDubWgYCh",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.2-2025-12-11_low",
      "score": 0.7888888888888889,
      "bestScore": 0.7888888888888889,
      "standardError": 0.05006659426035015,
      "startedAt": "2025-12-11T20:24:25.981Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F5Hcw3Td96qF66quWrDMKJB.eval",
      "sourceId": "5Hcw3Td96qF66quWrDMKJB",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Qwen3-235B-A22B-Thinking-2507",
      "score": 0.8666666666666667,
      "bestScore": 0.8666666666666667,
      "standardError": 0.051247074319053836,
      "startedAt": "2025-12-10T18:03:27.753Z",
      "releaseDate": "2025-07-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXWpHpVPCaVjcwPpdpXuuGH.eval",
      "sourceId": "XWpHpVPCaVjcwPpdpXuuGH",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.1-2025-11-13_low",
      "score": 0.6388888888888888,
      "bestScore": 0.6388888888888888,
      "standardError": 0.058821311521534046,
      "startedAt": "2025-11-25T15:18:46.210Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6J9aPwieMfq2eSse8ZknBW.eval",
      "sourceId": "6J9aPwieMfq2eSse8ZknBW",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-5-20251101",
      "score": 0.48055555555555557,
      "bestScore": 0.48055555555555557,
      "standardError": 0.06405336755004033,
      "startedAt": "2025-11-24T20:47:49.364Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FnSb2focDRMyPJXErk3MV94.eval",
      "sourceId": "nSb2focDRMyPJXErk3MV94",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-5-20251101_32K",
      "score": 0.8611111111111112,
      "bestScore": 0.8611111111111112,
      "standardError": 0.042644255740946095,
      "startedAt": "2025-11-24T20:14:50.875Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FikUWdu6HJ2qRgZynfzDCFe.eval",
      "sourceId": "ikUWdu6HJ2qRgZynfzDCFe",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-5-20251101_16K",
      "score": 0.8166666666666667,
      "bestScore": 0.8166666666666667,
      "standardError": 0.05018903659106632,
      "startedAt": "2025-11-24T20:02:12.729Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSH5jr76ni3RnpXVMETMy2H.eval",
      "sourceId": "SH5jr76ni3RnpXVMETMy2H",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-3-pro-preview",
      "score": 0.9138888888888889,
      "bestScore": 0.9138888888888889,
      "standardError": 0.03658872186755414,
      "startedAt": "2025-11-19T15:16:13.231Z",
      "releaseDate": "2025-11-18",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F4QM4h7ndQiDRL2anjKeddK.eval",
      "sourceId": "4QM4h7ndQiDRL2anjKeddK",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.1-2025-11-13_medium",
      "score": 0.8555555555555555,
      "bestScore": 0.8555555555555555,
      "standardError": 0.04387258359693737,
      "startedAt": "2025-11-17T21:28:27.888Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fk7inKYrGDT2LFcCiQhGfTP.eval",
      "sourceId": "k7inKYrGDT2LFcCiQhGfTP",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-2.5-pro",
      "score": 0.8416666666666667,
      "bestScore": 0.8472222222222222,
      "standardError": 0.048143196054711346,
      "startedAt": "2025-11-16T19:00:06.794Z",
      "releaseDate": "2025-06-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FiJp7PgYZa3PiXAV8UWd7MC.eval",
      "sourceId": "iJp7PgYZa3PiXAV8UWd7MC",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5.1-2025-11-13_high",
      "score": 0.8861111111111111,
      "bestScore": 0.8861111111111111,
      "standardError": 0.040479727089927024,
      "startedAt": "2025-11-13T19:52:15.595Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmYKKTdV27TRJv9FpUg7rkn.eval",
      "sourceId": "mYKKTdV27TRJv9FpUg7rkn",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "kimi-k2-thinking-turbo",
      "score": 0.8305555555555556,
      "bestScore": 0.8305555555555556,
      "standardError": 0.048192643258048534,
      "startedAt": "2025-11-13T14:21:18.345Z",
      "releaseDate": "2025-11-06",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F8BM7dCHaUZgcYJwZWiBhiQ.eval",
      "sourceId": "8BM7dCHaUZgcYJwZWiBhiQ",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5-nano-2025-08-07_high",
      "score": 0.8111111111111111,
      "bestScore": 0.8111111111111111,
      "standardError": 0.04626556218905063,
      "startedAt": "2025-10-31T09:55:38.159Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FY9YnaJQWpPFAPhe6jtKMhH.eval",
      "sourceId": "Y9YnaJQWpPFAPhe6jtKMhH",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5-mini-2025-08-07_high",
      "score": 0.8666666666666667,
      "bestScore": 0.8666666666666667,
      "standardError": 0.04286165199944133,
      "startedAt": "2025-10-30T09:59:54.926Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FR82gycJfm5jaCdh32tFkH4.eval",
      "sourceId": "R82gycJfm5jaCdh32tFkH4",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.9138888888888889,
      "bestScore": 0.9138888888888889,
      "standardError": 0.03658872186755414,
      "startedAt": "2025-10-29T22:29:41.753Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F3D8aaFBLZwQcnWHYKwvnmq.eval",
      "sourceId": "3D8aaFBLZwQcnWHYKwvnmq",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-5-20250929_59K",
      "score": 0.7777777777777778,
      "bestScore": 0.7777777777777778,
      "standardError": 0.06267511942419625,
      "startedAt": "2025-10-28T10:21:02.293Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fk5PDL6MpCC6P5eLAhodJiS.eval",
      "sourceId": "k5PDL6MpCC6P5eLAhodJiS",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-5-20250929_16K",
      "score": 0.7111111111111111,
      "bestScore": 0.7111111111111111,
      "standardError": 0.0683294324254051,
      "startedAt": "2025-10-28T09:57:34.696Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7QkPRbRPpRx8nJT2ZcqSx6.eval",
      "sourceId": "7QkPRbRPpRx8nJT2ZcqSx6",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-4-0613",
      "score": 0.011111111111111112,
      "bestScore": 0.011111111111111112,
      "standardError": 0.005362818328685681,
      "startedAt": "2025-10-23T16:09:26.069Z",
      "releaseDate": "2023-06-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhGVxg4EhSSJSCdH88m9soq.eval",
      "sourceId": "hGVxg4EhSSJSCdH88m9soq",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-4-0314",
      "score": 0.005555555555555556,
      "bestScore": 0.005555555555555556,
      "standardError": 0.003883473863441843,
      "startedAt": "2025-10-22T16:49:17.118Z",
      "releaseDate": "2023-03-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcLTgetxaaTCZoA9vwY4SdL.eval",
      "sourceId": "cLTgetxaaTCZoA9vwY4SdL",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-haiku-4-5-20251001_32K",
      "score": 0.6666666666666666,
      "bestScore": 0.6666666666666666,
      "standardError": 0.07106690545187014,
      "startedAt": "2025-10-22T10:31:07.212Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FneZvEzLi6yeL9z5KjGef7H.eval",
      "sourceId": "neZvEzLi6yeL9z5KjGef7H",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-5-20250929_32K",
      "score": 0.7777777777777778,
      "bestScore": 0.7777777777777778,
      "standardError": 0.06267511942419625,
      "startedAt": "2025-10-21T12:33:20.414Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FkqTyhmHyT9irTAWk3xBt2x.eval",
      "sourceId": "kqTyhmHyT9irTAWk3xBt2x",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-haiku-4-5-20251001",
      "score": 0.35833333333333334,
      "bestScore": 0.35833333333333334,
      "standardError": 0.06238626014232183,
      "startedAt": "2025-10-16T11:58:20.630Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWf2kwcXu8YXNgzDCxChtvj.eval",
      "sourceId": "Wf2kwcXu8YXNgzDCxChtvj",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen3-max-2025-09-23",
      "score": 0.7333333333333333,
      "bestScore": 0.7333333333333333,
      "standardError": 0.059139451700829515,
      "startedAt": "2025-10-06T14:32:01.826Z",
      "releaseDate": "2025-09-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FULeuPtwp3ZoyuXQYuuQ82i.eval",
      "sourceId": "ULeuPtwp3ZoyuXQYuuQ82i",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-5-20250929",
      "score": 0.35555555555555557,
      "bestScore": 0.35555555555555557,
      "standardError": 0.0721639236343101,
      "startedAt": "2025-09-29T18:39:45.629Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdEVgttSw8u6Tt2RsPWTc3C.eval",
      "sourceId": "dEVgttSw8u6Tt2RsPWTc3C",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5-2025-08-07_medium",
      "score": 0.8722222222222222,
      "bestScore": 0.8722222222222222,
      "standardError": 0.03882118926889545,
      "startedAt": "2025-08-07T21:58:31.032Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FctsGxm6WkzCGEqiAbJR3kk.eval",
      "sourceId": "ctsGxm6WkzCGEqiAbJR3kk",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5-mini-2025-08-07_medium",
      "score": 0.7833333333333333,
      "bestScore": 0.7833333333333333,
      "standardError": 0.04790431329560243,
      "startedAt": "2025-08-07T19:34:33.118Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F65DJZN2XzPw4RGpa4TCfA5.eval",
      "sourceId": "65DJZN2XzPw4RGpa4TCfA5",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-5-nano-2025-08-07_medium",
      "score": 0.7416666666666667,
      "bestScore": 0.7416666666666667,
      "standardError": 0.05231388366621641,
      "startedAt": "2025-08-07T19:34:30.191Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXRFS8R4WV7q6RMv5keEAdM.eval",
      "sourceId": "XRFS8R4WV7q6RMv5keEAdM",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-1-20250805_27K",
      "score": 0.6888888888888889,
      "bestScore": 0.6888888888888889,
      "standardError": 0.06979205927323111,
      "startedAt": "2025-08-05T16:53:30.846Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FUH9RY8MCAotDgFocHKA4j6.eval",
      "sourceId": "UH9RY8MCAotDgFocHKA4j6",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-1-20250805_16K",
      "score": 0.6444444444444445,
      "bestScore": 0.6444444444444445,
      "standardError": 0.0721639236343101,
      "startedAt": "2025-08-05T16:51:45.842Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fkf9vy9sMjMBBswrX6NMzce.eval",
      "sourceId": "kf9vy9sMjMBBswrX6NMzce",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-1-20250805",
      "score": 0.4,
      "bestScore": 0.4,
      "standardError": 0.07385489458759964,
      "startedAt": "2025-08-05T16:45:26.245Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRaiKuDSzUJYDVMxn4jfzNC.eval",
      "sourceId": "RaiKuDSzUJYDVMxn4jfzNC",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-4o-mini-2024-07-18",
      "score": 0.06944444444444445,
      "bestScore": 0.06944444444444445,
      "standardError": 0.0273963714244758,
      "startedAt": "2025-07-30T15:47:06.669Z",
      "releaseDate": "2024-07-18",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMPJkYPEMsQrowqtecK3mmz.eval",
      "sourceId": "MPJkYPEMsQrowqtecK3mmz",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "DeepSeek-R1-0528",
      "score": 0.6638888888888889,
      "bestScore": 0.6638888888888889,
      "standardError": 0.04911177955575991,
      "startedAt": "2025-05-29T10:59:10.779Z",
      "releaseDate": "2025-05-28",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbBV5wCrrZtihrgxEsg9jPD.eval",
      "sourceId": "bBV5wCrrZtihrgxEsg9jPD",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-20250514_27K",
      "score": 0.6444444444444445,
      "bestScore": 0.6444444444444445,
      "standardError": 0.0721639236343101,
      "startedAt": "2025-05-28T18:42:14.809Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVrRccga4uRVuCjNp637MeY.eval",
      "sourceId": "VrRccga4uRVuCjNp637MeY",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-20250514_59K",
      "score": 0.6888888888888889,
      "bestScore": 0.6888888888888889,
      "standardError": 0.06979205927323111,
      "startedAt": "2025-05-23T21:01:53.668Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fg8qaq3s2TJqfvrskHYmapM.eval",
      "sourceId": "g8qaq3s2TJqfvrskHYmapM",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-20250514_32K",
      "score": 0.7111111111111111,
      "bestScore": 0.7111111111111111,
      "standardError": 0.06832943242540508,
      "startedAt": "2025-05-22T23:06:11.220Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFZHzYjwAQexVykNQTnRtwT.eval",
      "sourceId": "FZHzYjwAQexVykNQTnRtwT",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-20250514_16K",
      "score": 0.6,
      "bestScore": 0.6,
      "standardError": 0.07385489458759964,
      "startedAt": "2025-05-22T22:54:06.990Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXz2F28L5k8HFNfY7Lo8Ggw.eval",
      "sourceId": "Xz2F28L5k8HFNfY7Lo8Ggw",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-20250514_16K",
      "score": 0.5333333333333333,
      "bestScore": 0.5333333333333333,
      "standardError": 0.07521014330903548,
      "startedAt": "2025-05-22T22:54:05.950Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FnDrnRwsADSyj8U6BD4in4r.eval",
      "sourceId": "nDrnRwsADSyj8U6BD4in4r",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-opus-4-20250514",
      "score": 0.4222222222222222,
      "bestScore": 0.4222222222222222,
      "standardError": 0.07446027270295805,
      "startedAt": "2025-05-22T17:26:48.979Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FaCbfcxGpLpPc9XkZNcuF4o.eval",
      "sourceId": "aCbfcxGpLpPc9XkZNcuF4o",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-sonnet-4-20250514",
      "score": 0.28888888888888886,
      "bestScore": 0.28888888888888886,
      "standardError": 0.06832943242540508,
      "startedAt": "2025-05-22T17:26:46.358Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FUa8sCkk3R3mBLWeSnnPG7a.eval",
      "sourceId": "Ua8sCkk3R3mBLWeSnnPG7a",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-2.5-flash-preview-05-20",
      "score": 0.7083333333333334,
      "bestScore": 0.7083333333333334,
      "standardError": 0.05715807427254223,
      "startedAt": "2025-05-20T21:22:01.376Z",
      "releaseDate": "2025-05-20",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9e62UjoHQbcZD2xjgfRkdK.eval",
      "sourceId": "9e62UjoHQbcZD2xjgfRkdK",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "mistral-medium-2505",
      "score": 0.32222222222222224,
      "bestScore": 0.32222222222222224,
      "standardError": 0.05965611798482264,
      "startedAt": "2025-05-07T17:32:21.604Z",
      "releaseDate": "2025-05-07",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQWw65xeTn8x42SymMsVREN.eval",
      "sourceId": "QWw65xeTn8x42SymMsVREN",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-2.5-flash-preview-04-17",
      "score": 0.7305555555555555,
      "bestScore": 0.7305555555555555,
      "standardError": 0.05532781604070319,
      "startedAt": "2025-04-22T10:03:18.492Z",
      "releaseDate": "2025-04-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKpZYG5nvdJ6QYweLz6dn5a.eval",
      "sourceId": "KpZYG5nvdJ6QYweLz6dn5a",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o4-mini-2025-04-16_high",
      "score": 0.8166666666666667,
      "bestScore": 0.8166666666666667,
      "standardError": 0.046939139620180814,
      "startedAt": "2025-04-16T18:17:08.217Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTVsxkJ8ZyfMVmVx88Qq3KN.eval",
      "sourceId": "TVsxkJ8ZyfMVmVx88Qq3KN",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o3-2025-04-16_high",
      "score": 0.8388888888888889,
      "bestScore": 0.8388888888888889,
      "standardError": 0.04381258550587166,
      "startedAt": "2025-04-16T18:17:05.608Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6anLU9CfBBNQkWyJG2GDCa.eval",
      "sourceId": "6anLU9CfBBNQkWyJG2GDCa",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-4.1-nano-2025-04-14",
      "score": 0.28888888888888886,
      "bestScore": 0.28888888888888886,
      "standardError": 0.05970019560922746,
      "startedAt": "2025-04-14T22:50:00.937Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F3qLHwibDhgtWEbKGwxUYVa.eval",
      "sourceId": "3qLHwibDhgtWEbKGwxUYVa",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-4.1-mini-2025-04-14",
      "score": 0.44722222222222224,
      "bestScore": 0.44722222222222224,
      "standardError": 0.06148017573965347,
      "startedAt": "2025-04-14T22:49:56.953Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcXbbdQDW2jeYassmRuZYpU.eval",
      "sourceId": "cXbbdQDW2jeYassmRuZYpU",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-4.1-2025-04-14",
      "score": 0.38333333333333336,
      "bestScore": 0.38333333333333336,
      "standardError": 0.06324056203789373,
      "startedAt": "2025-04-14T18:34:29.241Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVAQtARXUPTpdxGKMekUSbX.eval",
      "sourceId": "VAQtARXUPTpdxGKMekUSbX",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "grok-3-mini-beta_low",
      "score": 0.6222222222222222,
      "bestScore": 0.6222222222222222,
      "standardError": 0.07309112127323451,
      "startedAt": "2025-04-10T20:30:32.715Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTTmhagb2ovbi7GJnDxEfrj.eval",
      "sourceId": "TTmhagb2ovbi7GJnDxEfrj",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "grok-3-mini-beta_high",
      "score": 0.7777777777777778,
      "bestScore": 0.7777777777777778,
      "standardError": 0.06267511942419625,
      "startedAt": "2025-04-10T18:02:28.291Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfPtize44vDPcDAzMFuAurw.eval",
      "sourceId": "fPtize44vDPcDAzMFuAurw",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "grok-3-beta",
      "score": 0.5555555555555556,
      "bestScore": 0.5555555555555556,
      "standardError": 0.07491109582924914,
      "startedAt": "2025-04-10T17:55:19.398Z",
      "releaseDate": "2025-04-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDKTzh9kSoipZGkDsC4NGf3.eval",
      "sourceId": "DKTzh9kSoipZGkDsC4NGf3",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Llama-4-Maverick-17B-128E-Instruct-FP8",
      "score": 0.20555555555555555,
      "bestScore": 0.20555555555555555,
      "standardError": 0.04948528559541873,
      "startedAt": "2025-04-08T09:58:55.652Z",
      "releaseDate": "2025-04-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGJEpPThqUxXJBSbvkuXVnw.eval",
      "sourceId": "GJEpPThqUxXJBSbvkuXVnw",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Llama-4-Scout-17B-16E-Instruct",
      "score": 0.07777777777777778,
      "bestScore": 0.07777777777777778,
      "standardError": 0.030069597533844094,
      "startedAt": "2025-04-08T09:43:28.090Z",
      "releaseDate": "2025-04-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FT3X5DeC47iqCaoRYXu5k8L.eval",
      "sourceId": "T3X5DeC47iqCaoRYXu5k8L",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen-turbo-2024-11-01",
      "score": 0.06111111111111111,
      "bestScore": 0.06111111111111111,
      "standardError": 0.022340279838327228,
      "startedAt": "2025-04-07T19:18:53.147Z",
      "releaseDate": "2024-11-01",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FG5gxbTH33S7LXA29o5fgmv.eval",
      "sourceId": "G5gxbTH33S7LXA29o5fgmv",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen-plus-2025-01-25",
      "score": 0.17777777777777778,
      "bestScore": 0.17777777777777778,
      "standardError": 0.04360797453817489,
      "startedAt": "2025-04-07T19:09:43.970Z",
      "releaseDate": "2025-01-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FAULNJG2C75JBP6uaiAgj3P.eval",
      "sourceId": "AULNJG2C75JBP6uaiAgj3P",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen-max-2025-01-25",
      "score": 0.16111111111111112,
      "bestScore": 0.16111111111111112,
      "standardError": 0.04102194808229969,
      "startedAt": "2025-04-01T17:11:08.937Z",
      "releaseDate": "2025-01-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPZt4Yb9bpXVPDoPi4uTxwK.eval",
      "sourceId": "PZt4Yb9bpXVPDoPi4uTxwK",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Hermes-2-Theta-Llama-3-70B",
      "score": 0.025,
      "bestScore": 0.025,
      "standardError": 0.012309149097933271,
      "startedAt": "2025-04-01T15:03:10.644Z",
      "releaseDate": "2024-06-20",
      "organization": "Nous Research,Arcee AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F63MnE4e4AXvbQ585MaWTFv.eval",
      "sourceId": "63MnE4e4AXvbQ585MaWTFv",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "DeepSeek-V3-0324",
      "score": 0.37777777777777777,
      "bestScore": 0.37777777777777777,
      "standardError": 0.062372203800719374,
      "startedAt": "2025-04-01T14:33:01.704Z",
      "releaseDate": "2025-03-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVpsMexBEhXkPYJBe7v6Fk9.eval",
      "sourceId": "VpsMexBEhXkPYJBe7v6Fk9",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "mistral-small-2501",
      "score": 0.05277777777777778,
      "bestScore": 0.05277777777777778,
      "standardError": 0.02589584594604903,
      "startedAt": "2025-03-18T16:39:25.325Z",
      "releaseDate": "2025-01-25",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fb4vEwq8UVmGmAKh4eqp6G3.eval",
      "sourceId": "b4vEwq8UVmGmAKh4eqp6G3",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "mistral-small-2503",
      "score": 0.058333333333333334,
      "bestScore": 0.058333333333333334,
      "standardError": 0.0226830377599695,
      "startedAt": "2025-03-18T16:08:02.049Z",
      "releaseDate": "2025-03-17",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FJxqtHWrwCtYoQYBB3jVqB9.eval",
      "sourceId": "JxqtHWrwCtYoQYBB3jVqB9",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-3-7-sonnet-20250219_64K",
      "score": 0.5777777777777777,
      "bestScore": 0.5777777777777777,
      "standardError": 0.07446027270295805,
      "startedAt": "2025-03-13T21:20:00.444Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQW3FUBn5qtAFEX4oQuzEAV.eval",
      "sourceId": "QW3FUBn5qtAFEX4oQuzEAV",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-3-7-sonnet-20250219_32K",
      "score": 0.5333333333333333,
      "bestScore": 0.5333333333333333,
      "standardError": 0.07521014330903548,
      "startedAt": "2025-03-12T19:36:44.645Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdSuGgEbnQErengjEboPn62.eval",
      "sourceId": "dSuGgEbnQErengjEboPn62",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Llama-3.1-Tulu-3-70B-DPO",
      "score": 0.044444444444444446,
      "bestScore": 0.044444444444444446,
      "standardError": 0.01827278031830863,
      "startedAt": "2025-03-07T23:13:12.814Z",
      "releaseDate": "2024-11-21",
      "organization": "Allen Institute for AI,University of Washington",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFGcofc566CradiNmHUGqVB.eval",
      "sourceId": "FGcofc566CradiNmHUGqVB",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemma-2-27b-it",
      "score": 0.013888888888888888,
      "bestScore": 0.013888888888888888,
      "standardError": 0.00907862525871088,
      "startedAt": "2025-03-07T23:11:28.104Z",
      "releaseDate": "2024-06-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjD6rJdvfNPErKEz9qkpssj.eval",
      "sourceId": "jD6rJdvfNPErKEz9qkpssj",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemma-2-9b-it",
      "score": 0.005555555555555556,
      "bestScore": 0.005555555555555556,
      "standardError": 0.003883473863441843,
      "startedAt": "2025-03-07T23:11:27.610Z",
      "releaseDate": "2024-06-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWNE7cveVN8rsi7DmsFsFVy.eval",
      "sourceId": "WNE7cveVN8rsi7DmsFsFVy",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "DeepSeek-R1-Distill-Llama-70B",
      "score": 0.5138888888888888,
      "bestScore": 0.5138888888888888,
      "standardError": 0.060540263104731575,
      "startedAt": "2025-03-07T23:03:59.002Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F89zcekqEYnE8zhwG4pR5Nu.eval",
      "sourceId": "89zcekqEYnE8zhwG4pR5Nu",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-2.1",
      "score": 0.019444444444444445,
      "bestScore": 0.019444444444444445,
      "standardError": 0.010477515876628662,
      "startedAt": "2025-03-07T22:18:53.415Z",
      "releaseDate": "2023-11-21",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2vg9qw9pd8zSyRMu9YNiVm.eval",
      "sourceId": "2vg9qw9pd8zSyRMu9YNiVm",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-2.0",
      "score": 0.025,
      "bestScore": 0.025,
      "standardError": 0.015179897658722631,
      "startedAt": "2025-03-07T22:18:52.824Z",
      "releaseDate": "2023-07-11",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7LLTmqs4LC5GLVfJEqJWSC.eval",
      "sourceId": "7LLTmqs4LC5GLVfJEqJWSC",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-2.0-flash-thinking-exp-01-21",
      "score": 0.5777777777777777,
      "bestScore": 0.5777777777777777,
      "standardError": 0.07446027270295805,
      "startedAt": "2025-03-07T22:18:52.419Z",
      "releaseDate": "2025-01-21",
      "organization": "Google DeepMind,Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcXz2gsCmGYSvQ6Ey4LDiN7.eval",
      "sourceId": "cXz2gsCmGYSvQ6Ey4LDiN7",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o1-preview-2024-09-12",
      "score": 0.3111111111111111,
      "bestScore": 0.3111111111111111,
      "standardError": 0.0697920592732311,
      "startedAt": "2025-03-07T22:14:04.083Z",
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCprjaj9JQ9MY5NCXqeQR5H.eval",
      "sourceId": "Cprjaj9JQ9MY5NCXqeQR5H",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o1-mini-2024-09-12_high",
      "score": 0.46944444444444444,
      "bestScore": 0.46944444444444444,
      "standardError": 0.06456873681038247,
      "startedAt": "2025-03-06T21:47:29.432Z",
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9zhPZLvNeUPQ7JBSpTwKPp.eval",
      "sourceId": "9zhPZLvNeUPQ7JBSpTwKPp",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o1-mini-2024-09-12_medium",
      "score": 0.44722222222222224,
      "bestScore": 0.44722222222222224,
      "standardError": 0.06312676751311953,
      "startedAt": "2025-03-06T21:47:28.091Z",
      "releaseDate": "2024-09-12",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQUTfpiKjphedrKg8H6fGzF.eval",
      "sourceId": "QUTfpiKjphedrKg8H6fGzF",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-4.5-preview-2025-02-27",
      "score": 0.37777777777777777,
      "bestScore": 0.37777777777777777,
      "standardError": 0.07309112127323451,
      "startedAt": "2025-02-28T09:17:11.571Z",
      "releaseDate": "2025-02-27",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9DCJMt7YsTMDenTdX4Db6V.eval",
      "sourceId": "9DCJMt7YsTMDenTdX4Db6V",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o3-mini-2025-01-31_high",
      "score": 0.7694444444444445,
      "bestScore": 0.7694444444444445,
      "standardError": 0.05239762076828802,
      "startedAt": "2025-02-27T14:29:45.288Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMjx6vYbgAb2DHXK8xcupX8.eval",
      "sourceId": "Mjx6vYbgAb2DHXK8xcupX8",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-4-turbo-2024-04-09",
      "score": 0.06666666666666667,
      "bestScore": 0.06666666666666667,
      "standardError": 0.024034381938205007,
      "startedAt": "2025-02-27T14:29:44.767Z",
      "releaseDate": "2024-04-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FMZABVPEY8VCPAyCZ4iir68.eval",
      "sourceId": "MZABVPEY8VCPAyCZ4iir68",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o1-2024-12-17_medium",
      "score": 0.7333333333333333,
      "bestScore": 0.7333333333333333,
      "standardError": 0.06666666666666667,
      "startedAt": "2025-02-27T14:16:13.444Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKvcqycaJpCQMBigiEosyzC.eval",
      "sourceId": "KvcqycaJpCQMBigiEosyzC",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-3-7-sonnet-20250219_16K",
      "score": 0.4666666666666667,
      "bestScore": 0.4666666666666667,
      "standardError": 0.07521014330903548,
      "startedAt": "2025-02-26T16:56:46.931Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F64LAJymM6eXDkRZb92Mdg8.eval",
      "sourceId": "64LAJymM6eXDkRZb92Mdg8",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "DeepSeek-R1",
      "score": 0.5333333333333333,
      "bestScore": 0.5333333333333333,
      "standardError": 0.07521014330903548,
      "startedAt": "2025-02-26T16:56:45.594Z",
      "releaseDate": "2025-01-20",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbBY5VZzpDkmTBdSSvu5Ups.eval",
      "sourceId": "bBY5VZzpDkmTBdSSvu5Ups",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Meta-Llama-3-70B-Instruct",
      "score": 0.043055555555555555,
      "bestScore": 0.043055555555555555,
      "standardError": 0.02192429710598427,
      "startedAt": "2025-02-25T14:38:33.050Z",
      "releaseDate": "2024-04-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6RD3A6J432TZzxrKYRxph2.eval",
      "sourceId": "6RD3A6J432TZzxrKYRxph2",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-1.5-pro-001",
      "score": 0.06805555555555555,
      "bestScore": 0.06805555555555555,
      "standardError": 0.01893034295514878,
      "startedAt": "2025-02-25T14:38:32.379Z",
      "releaseDate": "2024-05-14",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F3S6fjGAMKWKK3zcdxkmh9H.eval",
      "sourceId": "3S6fjGAMKWKK3zcdxkmh9H",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "grok-2-1212",
      "score": 0.11527777777777778,
      "bestScore": 0.11527777777777778,
      "standardError": 0.03382424700930314,
      "startedAt": "2025-02-25T14:38:30.971Z",
      "releaseDate": "2024-12-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPpSVDb67Ucbxpcd8JQv5ii.eval",
      "sourceId": "PpSVDb67Ucbxpcd8JQv5ii",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Llama-3.3-70B-Instruct",
      "score": 0.05138888888888889,
      "bestScore": 0.05138888888888889,
      "standardError": 0.023611111111111107,
      "startedAt": "2025-02-25T14:05:10.812Z",
      "releaseDate": "2024-12-06",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZjsXQpZ7SsejXDo52aetZZ.eval",
      "sourceId": "ZjsXQpZ7SsejXDo52aetZZ",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Llama-3.2-90B-Vision-Instruct",
      "score": 0.02638888888888889,
      "bestScore": 0.02638888888888889,
      "standardError": 0.00941990275434065,
      "startedAt": "2025-02-25T14:05:10.587Z",
      "releaseDate": "2024-09-24",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLHRL6nQJZXGJmYppR7kbGw.eval",
      "sourceId": "LHRL6nQJZXGJmYppR7kbGw",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Llama-2-70b-chat-hf",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2025-02-25T14:05:09.775Z",
      "releaseDate": "2023-07-18",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FNLk2oE4rzSFihPzF95peHr.eval",
      "sourceId": "NLk2oE4rzSFihPzF95peHr",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-3-5-sonnet-20240620",
      "score": 0.06527777777777778,
      "bestScore": 0.06527777777777778,
      "standardError": 0.022293131815660016,
      "startedAt": "2025-02-25T14:05:09.497Z",
      "releaseDate": "2024-06-20",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2xNPhKLsiWLHQBLLDGU4VJ.eval",
      "sourceId": "2xNPhKLsiWLHQBLLDGU4VJ",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-1.5-pro-002",
      "score": 0.23055555555555557,
      "bestScore": 0.23055555555555557,
      "standardError": 0.05186782011216038,
      "startedAt": "2025-02-25T14:05:08.808Z",
      "releaseDate": "2024-09-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfhCdyX9uUS4WPtjLiVdBdK.eval",
      "sourceId": "fhCdyX9uUS4WPtjLiVdBdK",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-4o-2024-08-06",
      "score": 0.06388888888888888,
      "bestScore": 0.06388888888888888,
      "standardError": 0.025630380139261548,
      "startedAt": "2025-02-25T14:05:08.205Z",
      "releaseDate": "2024-08-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmBUKqiVQsqXd3qmdtKP5UP.eval",
      "sourceId": "mBUKqiVQsqXd3qmdtKP5UP",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-1.5-flash-002",
      "score": 0.1625,
      "bestScore": 0.1625,
      "standardError": 0.04111375571619442,
      "startedAt": "2025-02-25T14:05:07.974Z",
      "releaseDate": "2024-09-24",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQdGZ2m52PWt3EkLPbcoRwz.eval",
      "sourceId": "QdGZ2m52PWt3EkLPbcoRwz",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen2.5-32b-instruct",
      "score": 0.07361111111111111,
      "bestScore": 0.07361111111111111,
      "standardError": 0.021420595022864277,
      "startedAt": "2025-02-25T14:05:07.648Z",
      "releaseDate": "2024-09-17",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FNRCNnCCL5aQKsZf6cKC69Z.eval",
      "sourceId": "NRCNnCCL5aQKsZf6cKC69Z",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "mistral-large-2402",
      "score": 0.019444444444444445,
      "bestScore": 0.019444444444444445,
      "standardError": 0.00948945749721607,
      "startedAt": "2025-02-25T14:05:07.542Z",
      "releaseDate": "2024-02-26",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZVHvk3GTPUNTgjc5w3vkhW.eval",
      "sourceId": "ZVHvk3GTPUNTgjc5w3vkhW",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-1.5-flash-8b-001",
      "score": 0.04583333333333333,
      "bestScore": 0.04583333333333333,
      "standardError": 0.020100756305184247,
      "startedAt": "2025-02-25T14:05:06.918Z",
      "releaseDate": "2024-10-03",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjGY5Ee5P9dyStA4Fw5Pa6q.eval",
      "sourceId": "jGY5Ee5P9dyStA4Fw5Pa6q",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "mistral-large-2411",
      "score": 0.07777777777777778,
      "bestScore": 0.07777777777777778,
      "standardError": 0.024662778950157508,
      "startedAt": "2025-02-25T14:05:06.293Z",
      "releaseDate": "2024-11-18",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbeRGgJXTJggTi8PPEZNoZ5.eval",
      "sourceId": "beRGgJXTJggTi8PPEZNoZ5",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-3-sonnet-20240229",
      "score": 0.025,
      "bestScore": 0.025,
      "standardError": 0.01950297827553939,
      "startedAt": "2025-02-25T14:05:05.872Z",
      "releaseDate": "2024-02-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FAeRCEiG87bMFpxbxbmjCyn.eval",
      "sourceId": "AeRCEiG87bMFpxbxbmjCyn",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "DeepSeek-V3",
      "score": 0.15833333333333333,
      "bestScore": 0.15833333333333333,
      "standardError": 0.0433650088193118,
      "startedAt": "2025-02-25T14:05:05.855Z",
      "releaseDate": "2024-12-26",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7saSEhfVvXuz4yTcGy5u2y.eval",
      "sourceId": "7saSEhfVvXuz4yTcGy5u2y",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-1.0-pro-001",
      "score": 0.011111111111111112,
      "bestScore": 0.011111111111111112,
      "standardError": 0.006054026310473157,
      "startedAt": "2025-02-25T14:05:05.526Z",
      "releaseDate": "2023-12-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FNpJvYQDcmwbGgS5MzZUPUg.eval",
      "sourceId": "NpJvYQDcmwbGgS5MzZUPUg",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Llama-3.1-405B-Instruct",
      "score": 0.09722222222222222,
      "bestScore": 0.09722222222222222,
      "standardError": 0.03218644609162339,
      "startedAt": "2025-02-25T14:05:05.407Z",
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2fQJP3brRApfBdQq7Zy3bm.eval",
      "sourceId": "2fQJP3brRApfBdQq7Zy3bm",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-3-7-sonnet-20250219",
      "score": 0.21944444444444444,
      "bestScore": 0.21944444444444444,
      "standardError": 0.04815168903711295,
      "startedAt": "2025-02-25T14:05:04.923Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FN2NcmgmHyGJ9HNmQe5yvcS.eval",
      "sourceId": "N2NcmgmHyGJ9HNmQe5yvcS",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-3-5-sonnet-20241022",
      "score": 0.08472222222222223,
      "bestScore": 0.08472222222222223,
      "standardError": 0.026363956855902478,
      "startedAt": "2025-02-25T14:05:03.625Z",
      "releaseDate": "2024-10-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgeVVaRWxuprKUqnW2VzvCa.eval",
      "sourceId": "geVVaRWxuprKUqnW2VzvCa",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "qwen2.5-72b-instruct",
      "score": 0.08055555555555556,
      "bestScore": 0.08055555555555556,
      "standardError": 0.026098213018967212,
      "startedAt": "2025-02-25T14:05:03.625Z",
      "releaseDate": "2024-09-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F67YbVQfJh7sd77KH3zFgTn.eval",
      "sourceId": "67YbVQfJh7sd77KH3zFgTn",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-4o-2024-05-13",
      "score": 0.0625,
      "bestScore": 0.0625,
      "standardError": 0.023249986423369555,
      "startedAt": "2025-02-25T14:05:03.623Z",
      "releaseDate": "2024-05-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9cCn5aW522cTRxJSE69crm.eval",
      "sourceId": "9cCn5aW522cTRxJSE69crm",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "Llama-3.1-70B-Instruct",
      "score": 0.03611111111111111,
      "bestScore": 0.03611111111111111,
      "standardError": 0.01691212350465157,
      "startedAt": "2025-02-25T14:05:03.270Z",
      "releaseDate": "2024-07-23",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FX2tQXGELXtEbc2VX6gARd7.eval",
      "sourceId": "X2tQXGELXtEbc2VX6gARd7",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-3-5-haiku-20241022",
      "score": 0.043055555555555555,
      "bestScore": 0.043055555555555555,
      "standardError": 0.016237638827815495,
      "startedAt": "2025-02-25T14:05:03.257Z",
      "releaseDate": "2024-10-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcqhrR5KnwpcuAVpHkZGQFd.eval",
      "sourceId": "cqhrR5KnwpcuAVpHkZGQFd",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-1.5-flash-001",
      "score": 0.03888888888888889,
      "bestScore": 0.03888888888888889,
      "standardError": 0.014363726487520755,
      "startedAt": "2025-02-25T14:05:02.246Z",
      "releaseDate": "2024-05-23",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FU9GVBhYjRzU55YVjna5mMT.eval",
      "sourceId": "U9GVBhYjRzU55YVjna5mMT",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-3-haiku-20240307",
      "score": 0.018055555555555554,
      "bestScore": 0.018055555555555554,
      "standardError": 0.008098544298396251,
      "startedAt": "2025-02-25T14:05:02.017Z",
      "releaseDate": "2024-03-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FThMd6DNDTtyWxhPcb9Kq4y.eval",
      "sourceId": "ThMd6DNDTtyWxhPcb9Kq4y",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "claude-3-opus-20240229",
      "score": 0.04722222222222222,
      "bestScore": 0.04722222222222222,
      "standardError": 0.019579258477123922,
      "startedAt": "2025-02-25T14:05:00.461Z",
      "releaseDate": "2024-02-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmYb9MGQD39RS8VdDyMqujr.eval",
      "sourceId": "mYb9MGQD39RS8VdDyMqujr",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "o3-mini-2025-01-31_medium",
      "score": 0.6388888888888888,
      "bestScore": 0.6388888888888888,
      "standardError": 0.060377105937901555,
      "startedAt": "2025-02-25T13:33:25.508Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FajytUBY9bDPNNxuBfi5UBS.eval",
      "sourceId": "ajytUBY9bDPNNxuBfi5UBS",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "phi-4",
      "score": 0.1375,
      "bestScore": 0.1375,
      "standardError": 0.03698083382925635,
      "startedAt": "2025-02-25T13:19:42.221Z",
      "releaseDate": "2024-12-12",
      "organization": "Microsoft Research",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F5mmrBSQb7hVSJ9VYtNm46i.eval",
      "sourceId": "5mmrBSQb7hVSJ9VYtNm46i",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gpt-4o-2024-11-20",
      "score": 0.0625,
      "bestScore": 0.0625,
      "standardError": 0.024165360466267546,
      "startedAt": "2025-02-25T13:10:02.200Z",
      "releaseDate": "2024-11-20",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbuNVL2TWcdVgxAcMfVE8xJ.eval",
      "sourceId": "buNVL2TWcdVgxAcMfVE8xJ",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "gemini-2.0-flash-001",
      "score": 0.3111111111111111,
      "bestScore": 0.3111111111111111,
      "standardError": 0.06311218144454454,
      "startedAt": "2025-02-25T12:25:50.002Z",
      "releaseDate": "2025-02-05",
      "organization": "Google DeepMind,Google",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZnGSRrKbiEnuju4kk4sgVb.eval",
      "sourceId": "ZnGSRrKbiEnuju4kk4sgVb",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "mistral-large-2407",
      "score": 0.08472222222222223,
      "bestScore": 0.08472222222222223,
      "standardError": 0.024742642797709363,
      "startedAt": "2025-02-25T11:59:42.758Z",
      "releaseDate": "2024-07-24",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbHLfPfZj8HgD9Gef75NtrM.eval",
      "sourceId": "bHLfPfZj8HgD9Gef75NtrM",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "OTIS Mock AIME 2024-2025",
      "model": "grok-4-0709",
      "score": 0.84,
      "bestScore": 0.84,
      "standardError": 0.05,
      "startedAt": null,
      "releaseDate": "2025-07-09",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "cvTPRDCM38zSTn9Y3MUb9d",
      "sourceFile": "otis_mock_aime_2024_2025.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "glm-5.2_max",
      "score": 0.787,
      "bestScore": 0.787,
      "standardError": 0.018656911901656907,
      "startedAt": "2026-06-25T13:13:06.902Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F4AfhcmYVNrw6gM5u8CQsyy.eval",
      "sourceId": "4AfhcmYVNrw6gM5u8CQsyy",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "qwen3.7-max",
      "score": 0.7727272727272727,
      "bestScore": 0.7727272727272727,
      "standardError": 0.01906835447169652,
      "startedAt": "2026-06-18T17:48:14.587Z",
      "releaseDate": "2026-05-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F83hhDhcT9UUrZ2NiUbWSUU.eval",
      "sourceId": "83hhDhcT9UUrZ2NiUbWSUU",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "deepseek-v4-pro_max",
      "score": 0.7763975155279503,
      "bestScore": 0.7763975155279503,
      "standardError": 0.01897828052158487,
      "startedAt": "2026-06-18T17:43:54.039Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWUk3f44JhxtAJaSxi6gFZR.eval",
      "sourceId": "WUk3f44JhxtAJaSxi6gFZR",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gemini-3.5-flash_high",
      "score": 0.7933884297520661,
      "bestScore": 0.7933884297520661,
      "standardError": 0.018422420250424888,
      "startedAt": "2026-06-01T10:22:55.131Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdwRzFGTghYEy4HNZajJXoc.eval",
      "sourceId": "dwRzFGTghYEy4HNZajJXoc",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "qwen3.6-max-preview",
      "score": 0.7665289256198347,
      "bestScore": 0.7665289256198347,
      "standardError": 0.01924895909142592,
      "startedAt": "2026-05-28T10:24:53.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFkDMVLdEnJDR5PyUCT6BNQ.eval",
      "sourceId": "FkDMVLdEnJDR5PyUCT6BNQ",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "glm-5.1",
      "score": 0.7417,
      "bestScore": 0.7417,
      "standardError": 0.0199,
      "startedAt": "2026-05-15T10:07:59.494Z",
      "releaseDate": "2026-04-07",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "glm-5.1-swe-bench-frankenstein",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "qwen3.6-plus",
      "score": 0.5785123966942148,
      "bestScore": 0.5785123966942148,
      "standardError": 0.02246855652298021,
      "startedAt": "2026-05-14T17:28:26.954Z",
      "releaseDate": "2026-03-31",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F8HV4HP9EM5KuLSwzifuLye.eval",
      "sourceId": "8HV4HP9EM5KuLSwzifuLye",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "kimi-k2.6",
      "score": 0.7665289256198347,
      "bestScore": 0.7665289256198347,
      "standardError": 0.01924895909142592,
      "startedAt": "2026-05-08T10:21:24.110Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbUPtAK7spEwvmiC8pi6afc.eval",
      "sourceId": "bUPtAK7spEwvmiC8pi6afc",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gpt-5.5-pre-release_xhigh",
      "score": 0.8057851239669421,
      "bestScore": 0.8057851239669421,
      "standardError": 0.018000198869416815,
      "startedAt": "2026-04-24T13:52:36.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLbhPRqUkXrs5KJ7vyT4CR7.eval",
      "sourceId": "LbhPRqUkXrs5KJ7vyT4CR7",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "claude-opus-4-7_max",
      "score": 0.8347107438016529,
      "bestScore": 0.8347107438016529,
      "standardError": 0.016901169179529604,
      "startedAt": "2026-04-20T13:16:09.316Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FnCmGWWMBip2s9AVEZNYXDX.eval",
      "sourceId": "nCmGWWMBip2s9AVEZNYXDX",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gpt-5.4-2026-03-05_high",
      "score": 0.768595041322314,
      "bestScore": 0.768595041322314,
      "standardError": 0.019189407003691976,
      "startedAt": "2026-03-06T11:07:43.396Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fi9HTMbAxcwsrTf2w2FuJ6N.eval",
      "sourceId": "i9HTMbAxcwsrTf2w2FuJ6N",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gpt-5.3-codex_high",
      "score": 0.7479338842975206,
      "bestScore": 0.7479338842975206,
      "standardError": 0.019756739501473306,
      "startedAt": "2026-02-25T10:37:06.792Z",
      "releaseDate": "2026-02-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F6w5buMbgeEqtRoE55cLZVv.eval",
      "sourceId": "6w5buMbgeEqtRoE55cLZVv",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gemini-3.1-pro-preview-customtools",
      "score": 0.756198347107438,
      "bestScore": 0.756198347107438,
      "standardError": 0.019537213360116365,
      "startedAt": "2026-02-24T13:34:48.126Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F8QQQWDgmmEsmQVUJWcxx4P.eval",
      "sourceId": "8QQQWDgmmEsmQVUJWcxx4P",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "claude-sonnet-4-6",
      "score": 0.7520661157024794,
      "bestScore": 0.7520661157024794,
      "standardError": 0.01964818270139032,
      "startedAt": "2026-02-21T10:32:51.288Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKEdgiMmtJKHxgGiS82agYj.eval",
      "sourceId": "KEdgiMmtJKHxgGiS82agYj",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gemini-3-flash-preview",
      "score": 0.7541322314049587,
      "bestScore": 0.7541322314049587,
      "standardError": 0.019593002139508684,
      "startedAt": "2026-02-18T20:12:36.725Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjxoyMNmTnYirjLJN6wRNmG.eval",
      "sourceId": "jxoyMNmTnYirjLJN6wRNmG",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "claude-opus-4-6",
      "score": 0.7871900826446281,
      "bestScore": 0.7871900826446281,
      "standardError": 0.01862353730305288,
      "startedAt": "2026-02-18T20:07:57.705Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2hH2GAPUFaJR7pfVKhhicg.eval",
      "sourceId": "2hH2GAPUFaJR7pfVKhhicg",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gpt-5.1-2025-11-13_high",
      "score": 0.6797520661157025,
      "bestScore": 0.6797520661157025,
      "standardError": 0.02122974956319103,
      "startedAt": "2026-02-18T12:36:35.180Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FYoHtZpMNdsRE6qdj2K4Ucj.eval",
      "sourceId": "YoHtZpMNdsRE6qdj2K4Ucj",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "kimi-k2.5",
      "score": 0.737603305785124,
      "bestScore": 0.737603305785124,
      "standardError": 0.020017832064928274,
      "startedAt": "2026-02-17T14:43:08.990Z",
      "releaseDate": "2026-01-27",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F5wVjUb3yn6bdjPSfPEMuiZ.eval",
      "sourceId": "5wVjUb3yn6bdjPSfPEMuiZ",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gpt-5.1-2025-11-13_high",
      "score": 0.6590909090909091,
      "bestScore": 0.6590909090909091,
      "standardError": 0.021568422716496703,
      "startedAt": "2026-02-17T01:48:11.460Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FG5hTYKeE5buuWzwRQK7EgS.eval",
      "sourceId": "G5hTYKeE5buuWzwRQK7EgS",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "glm-5",
      "score": 0.7207792207792207,
      "bestScore": 0.7207792207792207,
      "standardError": 0.020894141602415234,
      "startedAt": "2026-02-15T15:40:10.848Z",
      "releaseDate": "2026-02-11",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FhKkJEu4ueLiNWsTtrJ78PS.eval",
      "sourceId": "hKkJEu4ueLiNWsTtrJ78PS",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gemini-2.5-pro",
      "score": 0.5755693581780539,
      "bestScore": 0.5755693581780539,
      "standardError": 0.02251275673889269,
      "startedAt": "2026-02-13T18:40:15.413Z",
      "releaseDate": "2025-06-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FKL724uHFGbiXj4o5PHyLn2.eval",
      "sourceId": "KL724uHFGbiXj4o5PHyLn2",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gemini-3-pro-preview",
      "score": 0.7293388429752066,
      "bestScore": 0.7293388429752066,
      "standardError": 0.02021641171153417,
      "startedAt": "2026-02-13T10:43:27.233Z",
      "releaseDate": "2025-11-18",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F5XdYZBG5Wy8DUMRqcYv4fn.eval",
      "sourceId": "5XdYZBG5Wy8DUMRqcYv4fn",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gpt-5.2-2025-12-11_high",
      "score": 0.737603305785124,
      "bestScore": 0.737603305785124,
      "standardError": 0.020017832064928277,
      "startedAt": "2026-02-12T13:34:00.865Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F9owdbpnWJ5pkQQC3GWKP5L.eval",
      "sourceId": "9owdbpnWJ5pkQQC3GWKP5L",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "o3-2025-04-16_medium",
      "score": 0.6231884057971014,
      "bestScore": 0.6231884057971014,
      "standardError": 0.022072335658124498,
      "startedAt": "2026-02-12T03:46:13.664Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FAKt7PcE8LQtoMWVuxwMidk.eval",
      "sourceId": "AKt7PcE8LQtoMWVuxwMidk",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "claude-opus-4-1-20250805",
      "score": 0.7334710743801653,
      "bestScore": 0.7334710743801653,
      "standardError": 0.020118245554677085,
      "startedAt": "2026-02-11T21:33:53.502Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcA6Mq67yaynLhY8F5kes3c.eval",
      "sourceId": "cA6Mq67yaynLhY8F5kes3c",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gpt-4o-2024-11-20",
      "score": 0.30991735537190085,
      "bestScore": 0.30991735537190085,
      "standardError": 0.021042625979133336,
      "startedAt": "2026-02-11T15:05:15.089Z",
      "releaseDate": "2024-11-20",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FPncXWJm7ZCY3qv982pKoRy.eval",
      "sourceId": "PncXWJm7ZCY3qv982pKoRy",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "claude-opus-4-6",
      "score": 0.756198347107438,
      "bestScore": 0.756198347107438,
      "standardError": 0.019537213360116365,
      "startedAt": "2026-02-10T22:23:09.876Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FaQiAg6mt8HPHXjKcJvax2d.eval",
      "sourceId": "aQiAg6mt8HPHXjKcJvax2d",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gpt-4.1-2025-04-14",
      "score": 0.48541666666666666,
      "bestScore": 0.48541666666666666,
      "standardError": 0.022835863746433965,
      "startedAt": "2026-02-08T17:41:23.879Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSheUt9cNpkSevy4r6Gsr3u.eval",
      "sourceId": "SheUt9cNpkSevy4r6Gsr3u",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.7355371900826446,
      "bestScore": 0.7355371900826446,
      "standardError": 0.02006832181710289,
      "startedAt": "2026-02-06T14:23:51.990Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FC85Bnci9xVzq5jmt8zNnbw.eval",
      "sourceId": "C85Bnci9xVzq5jmt8zNnbw",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "claude-opus-4-20250514",
      "score": 0.7066115702479339,
      "bestScore": 0.7066115702479339,
      "standardError": 0.02071754535128056,
      "startedAt": "2026-02-06T07:21:39.976Z",
      "releaseDate": "2025-05-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRbcvuVChvxK6SMQbPK8iRn.eval",
      "sourceId": "RbcvuVChvxK6SMQbPK8iRn",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "claude-opus-4-5-20251101",
      "score": 0.7665289256198347,
      "bestScore": 0.7665289256198347,
      "standardError": 0.01924895909142592,
      "startedAt": "2026-02-05T22:06:33.964Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmKQKxkpGDrFHBPBYR3mBrB.eval",
      "sourceId": "mKQKxkpGDrFHBPBYR3mBrB",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "claude-sonnet-4-5-20250929",
      "score": 0.7128099173553719,
      "bestScore": 0.7128099173553719,
      "standardError": 0.020587234937983246,
      "startedAt": "2026-02-05T16:52:49.622Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLti64yMEsBWSzgEMFCSVhp.eval",
      "sourceId": "Lti64yMEsBWSzgEMFCSVhp",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gpt-5-2025-08-07_medium",
      "score": 0.7148760330578512,
      "bestScore": 0.7148760330578512,
      "standardError": 0.020542754003902382,
      "startedAt": "2026-02-05T13:39:28.137Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFCCkMNVHcjLLUDeNFqTGRL.eval",
      "sourceId": "FCCkMNVHcjLLUDeNFqTGRL",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "claude-3-7-sonnet-20250219",
      "score": 0.609504132231405,
      "bestScore": 0.609504132231405,
      "standardError": 0.022198466971122266,
      "startedAt": "2026-02-04T18:05:21.852Z",
      "releaseDate": "2025-02-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FcnM8U6moeqrp74wnXRj2Jj.eval",
      "sourceId": "cnM8U6moeqrp74wnXRj2Jj",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SWE-Bench verified",
      "model": "gpt-5-mini-2025-08-07_medium",
      "score": 0.6466942148760331,
      "bestScore": 0.6466942148760331,
      "standardError": 0.021749601139671306,
      "startedAt": "2026-02-01T11:28:37.125Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7nvdQy3AwKtoay2FBvfST5.eval",
      "sourceId": "7nvdQy3AwKtoay2FBvfST5",
      "sourceFile": "swe_bench_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "Muse-Spark-Open-816B-A42B_xhigh",
      "score": 0.566,
      "bestScore": 0.566,
      "standardError": 0.01568087656637504,
      "startedAt": "2026-10-15T18:52:04.000Z",
      "releaseDate": "2026-10-15",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dNuxLdV4DWqfSvtTq6NnK8",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gemini-4-argon_high",
      "score": 0.76,
      "bestScore": 0.76,
      "standardError": 0.013512312258920845,
      "startedAt": "2026-10-13T17:15:14.000Z",
      "releaseDate": "2026-09-30",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iq4Y9tXpiyLP25yotySxJW",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "deepseek-v4.1-flash_max",
      "score": 0.429,
      "bestScore": 0.429,
      "standardError": 0.015658997547870177,
      "startedAt": "2026-10-13T12:59:33.000Z",
      "releaseDate": "2026-09-09",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "3JJ4ChNxMo8CGAh3HrKNrw",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-haiku-5-5_max",
      "score": 0.238,
      "bestScore": 0.238,
      "standardError": 0.013473586661967093,
      "startedAt": "2026-10-09T14:36:06.000Z",
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "NwnrUMtveqHDYB2Um4ycmA",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "mistral-large-4-0_high",
      "score": 0.2,
      "bestScore": 0.2,
      "standardError": 0.012655439943366657,
      "startedAt": "2026-10-09T14:12:09.000Z",
      "releaseDate": "2026-10-06",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "hvNNKdnzLvr9X8zVxWKcfk",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-6.1-sol_max",
      "score": 0.739,
      "bestScore": 0.739,
      "standardError": 0.013895037677965185,
      "startedAt": "2026-09-29T22:00:34.000Z",
      "releaseDate": "2026-09-29",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dfT5VD7cQug8XW6GQHyJUF",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-sonnet-5-5_max",
      "score": 0.465,
      "bestScore": 0.465,
      "standardError": 0.015780495050030086,
      "startedAt": "2026-09-29T15:15:14.000Z",
      "releaseDate": "2026-09-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GtJ5yd3ZPQ7HVcm6Q9ssQz",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-6-luna_max",
      "score": 0.414,
      "bestScore": 0.414,
      "standardError": 0.015583544104177553,
      "startedAt": "2026-09-22T22:14:05.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "HX9sRVWJtZiNvneALnVzs8",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "grok-4.7_xhigh",
      "score": 0.56,
      "bestScore": 0.56,
      "standardError": 0.015704987954361718,
      "startedAt": "2026-09-22T21:49:22.000Z",
      "releaseDate": "2026-09-21",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6izDCw5ygCLZJEKPS8jV5T",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-6-sol_max",
      "score": 0.607,
      "bestScore": 0.607,
      "standardError": 0.015452824654081525,
      "startedAt": "2026-09-22T19:59:24.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "donMeEKBZZdqpQsojKWJEm",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-opus-5-5_max",
      "score": 0.722,
      "bestScore": 0.722,
      "standardError": 0.014174516461485273,
      "startedAt": "2026-09-22T18:27:39.000Z",
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "CgLfiGcNwMPVzxr8GdNUAK",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "qwen3.8-max-0902_xhigh",
      "score": 0.473,
      "bestScore": 0.473,
      "standardError": 0.01579621855130273,
      "startedAt": "2026-09-02T18:04:58.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "nBcLJJomVoeg2zh5F8Mnsu",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gemini-3.8-flash_high",
      "score": 0.697,
      "bestScore": 0.697,
      "standardError": 0.014539683710535193,
      "startedAt": "2026-09-02T17:15:36.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gZ4bwozTKtpmZTqnqX3a33",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-fable-5-1_max",
      "score": 0.708,
      "bestScore": 0.708,
      "standardError": 0.014385511563477433,
      "startedAt": "2026-09-01T18:31:59.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ExyL3GN8s34tizrXV3VmS3",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-4o-mini-2024-07-18",
      "score": 0.083,
      "bestScore": 0.083,
      "standardError": 0.008728527206074758,
      "startedAt": "2026-08-31T22:02:07.000Z",
      "releaseDate": "2024-07-18",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDLr586bZUHpghAf8m3GRSm.eval",
      "sourceId": "DLr586bZUHpghAf8m3GRSm",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-4o-2024-08-06",
      "score": 0.26,
      "bestScore": 0.26,
      "standardError": 0.01387777332977422,
      "startedAt": "2026-08-31T22:02:02.000Z",
      "releaseDate": "2024-08-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbDFqfWxMG7gLDq4Mz8PRTP.eval",
      "sourceId": "bDFqfWxMG7gLDq4Mz8PRTP",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-4.1-nano-2025-04-14",
      "score": 0.06,
      "bestScore": 0.06,
      "standardError": 0.007513751157474971,
      "startedAt": "2026-08-31T22:01:58.000Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FA4JEeJR5LQfLR3Hb8z7kL2.eval",
      "sourceId": "A4JEeJR5LQfLR3Hb8z7kL2",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-4.1-mini-2025-04-14",
      "score": 0.127,
      "bestScore": 0.127,
      "standardError": 0.010534798620855644,
      "startedAt": "2026-08-31T22:01:51.000Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgYV4CJTjSd7VonfzMmnwcw.eval",
      "sourceId": "gYV4CJTjSd7VonfzMmnwcw",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-4.1-2025-04-14",
      "score": 0.311,
      "bestScore": 0.311,
      "standardError": 0.014645596385722692,
      "startedAt": "2026-08-31T22:01:48.000Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FavdYKfKUuY6EvfVdxftBE8.eval",
      "sourceId": "avdYKfKUuY6EvfVdxftBE8",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "o3-mini-2025-01-31_high",
      "score": 0.153,
      "bestScore": 0.153,
      "standardError": 0.011389500459665504,
      "startedAt": "2026-08-31T22:01:45.000Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FUCba4H3NNtUiWXMisWnh9g.eval",
      "sourceId": "UCba4H3NNtUiWXMisWnh9g",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "o1-2024-12-17_high",
      "score": 0.411,
      "bestScore": 0.411,
      "standardError": 0.01556667341859915,
      "startedAt": "2026-08-31T22:01:34.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLQfu4ngArcgMEyQZxuR7r4.eval",
      "sourceId": "LQfu4ngArcgMEyQZxuR7r4",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "muse-spark-1.1",
      "score": 0.5778894472361809,
      "bestScore": 0.5778894472361809,
      "standardError": 0.015665429072238852,
      "startedAt": "2026-08-31T19:21:58.000Z",
      "releaseDate": "2026-07-09",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbCdwpRXgZtvQLYC4wA9JXr.eval",
      "sourceId": "bCdwpRXgZtvQLYC4wA9JXr",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-3-opus-20240229",
      "score": 0.126,
      "bestScore": 0.126,
      "standardError": 0.010499249222407964,
      "startedAt": "2026-08-31T18:18:52.000Z",
      "releaseDate": "2024-02-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FNxVp4VF8uQbL5S6hx6HFRB.eval",
      "sourceId": "NxVp4VF8uQbL5S6hx6HFRB",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-6-astra_max",
      "score": 0.756,
      "bestScore": 0.756,
      "standardError": 0.013588548437881386,
      "startedAt": "2026-08-30T14:57:43.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PppwnX7EjXx2SeTcFN8Ujn",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "glm-5.3_max",
      "score": 0.41,
      "bestScore": 0.41,
      "standardError": 0.015560917136921659,
      "startedAt": "2026-08-28T17:00:06.000Z",
      "releaseDate": "2026-08-14",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTY8R8FV7zCwByEipkqyg2P.eval",
      "sourceId": "TY8R8FV7zCwByEipkqyg2P",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "qwen3.5-plus",
      "score": 0.2535070140280561,
      "bestScore": 0.2535070140280561,
      "standardError": 0.013777178222916077,
      "startedAt": "2026-08-27T21:29:42.000Z",
      "releaseDate": "2026-02-16",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FjrawnM2wHFbVfJFJsap4kS.eval",
      "sourceId": "jrawnM2wHFbVfJFJsap4kS",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "muse-spark-1.2_xhigh",
      "score": 0.6030150753768844,
      "bestScore": 0.6030150753768844,
      "standardError": 0.015518791563606454,
      "startedAt": "2026-08-27T19:36:49.000Z",
      "releaseDate": "2026-08-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmLLmVVvo6t23MM5yxq7ryW.eval",
      "sourceId": "mLLmVVvo6t23MM5yxq7ryW",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "Inkling_xhigh",
      "score": 0.403,
      "bestScore": 0.403,
      "standardError": 0.015518757419066513,
      "startedAt": "2026-08-27T19:36:45.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FbpnzC6CCi3hWuKCfqX4KEY.eval",
      "sourceId": "bpnzC6CCi3hWuKCfqX4KEY",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "Qwen3-235B-A22B-Thinking-2507",
      "score": 0.4044044044044044,
      "bestScore": 0.4044044044044044,
      "standardError": 0.015535255725079345,
      "startedAt": "2026-08-27T19:36:29.000Z",
      "releaseDate": "2025-07-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F4ACEC4x26HWRGiAepHGw43.eval",
      "sourceId": "4ACEC4x26HWRGiAepHGw43",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "glm-5.1",
      "score": 0.34,
      "bestScore": 0.34,
      "standardError": 0.014987482264363972,
      "startedAt": "2026-08-27T19:36:21.000Z",
      "releaseDate": "2026-04-07",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FR2fMFAtRXah864hMifdQGV.eval",
      "sourceId": "R2fMFAtRXah864hMifdQGV",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "inkling-small_xhigh",
      "score": 0.191,
      "bestScore": 0.191,
      "standardError": 0.012436787112179482,
      "startedAt": "2026-08-27T19:36:17.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FeJfbDKqeg9kNXFoxpeV7HG.eval",
      "sourceId": "eJfbDKqeg9kNXFoxpeV7HG",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "kimi-k3_max",
      "score": 0.506,
      "bestScore": 0.506,
      "standardError": 0.015818160898606836,
      "startedAt": "2026-08-27T19:36:12.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FmZLWt2JSYvJyGhMvji7CT8.eval",
      "sourceId": "mZLWt2JSYvJyGhMvji7CT8",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "glm-5.2_max",
      "score": 0.342,
      "bestScore": 0.342,
      "standardError": 0.015008706182121802,
      "startedAt": "2026-08-27T19:36:05.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FWgjFZzojpiv54XqUhvfuUG.eval",
      "sourceId": "WgjFZzojpiv54XqUhvfuUG",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "kimi-k2.7-code",
      "score": 0.365,
      "bestScore": 0.365,
      "standardError": 0.015231776226264846,
      "startedAt": "2026-08-27T19:36:03.000Z",
      "releaseDate": "2026-06-12",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F5AijYw5bFJtSuahnbzNo6Z.eval",
      "sourceId": "5AijYw5bFJtSuahnbzNo6Z",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "glm-4.7",
      "score": 0.322,
      "bestScore": 0.322,
      "standardError": 0.014782913600996744,
      "startedAt": "2026-08-27T19:36:02.000Z",
      "releaseDate": "2025-12-22",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FaYau6EG7CCaWVSyVq3vvQc.eval",
      "sourceId": "aYau6EG7CCaWVSyVq3vvQc",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "kimi-k2.5",
      "score": 0.343,
      "bestScore": 0.343,
      "standardError": 0.015019206922357007,
      "startedAt": "2026-08-27T19:35:45.000Z",
      "releaseDate": "2026-01-27",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRbXXg4JEbTSBnzw3CMSHnD.eval",
      "sourceId": "RbXXg4JEbTSBnzw3CMSHnD",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "qwen3.7-max",
      "score": 0.5576730190571715,
      "bestScore": 0.5576730190571715,
      "standardError": 0.015737359397180177,
      "startedAt": "2026-08-27T19:34:09.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDMjUrG6eZEqwcQ4SqUoViu.eval",
      "sourceId": "DMjUrG6eZEqwcQ4SqUoViu",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "qwen3.6-plus",
      "score": 0.44144144144144143,
      "bestScore": 0.44144144144144143,
      "standardError": 0.015718302238794238,
      "startedAt": "2026-08-27T19:34:06.000Z",
      "releaseDate": "2026-03-31",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F4LWNorP8Fy8V8doxirNFcH.eval",
      "sourceId": "4LWNorP8Fy8V8doxirNFcH",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "qwen3.6-max-preview",
      "score": 0.5200400801603207,
      "bestScore": 0.5200400801603207,
      "standardError": 0.015822434816352416,
      "startedAt": "2026-08-27T19:33:58.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FiyPiR8GUuyNNA6Fssziq32.eval",
      "sourceId": "iyPiR8GUuyNNA6Fssziq32",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "qwen3.6-flash",
      "score": 0.15915915915915915,
      "bestScore": 0.15915915915915915,
      "standardError": 0.011579970321213651,
      "startedAt": "2026-08-27T19:33:42.000Z",
      "releaseDate": "2026-04-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FFoMSS4FmH3VxpsvZiUJPCv.eval",
      "sourceId": "FoMSS4FmH3VxpsvZiUJPCv",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "qwen3.5-flash",
      "score": 0.2032032032032032,
      "bestScore": 0.2032032032032032,
      "standardError": 0.012737194890497593,
      "startedAt": "2026-08-27T19:32:07.000Z",
      "releaseDate": "2026-02-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FK8zTckSXLKLesoLBv24TWc.eval",
      "sourceId": "K8zTckSXLKLesoLBv24TWc",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "qwen3-max-2025-09-23",
      "score": 0.4874874874874875,
      "bestScore": 0.4874874874874875,
      "standardError": 0.01582226675546796,
      "startedAt": "2026-08-27T19:32:02.000Z",
      "releaseDate": "2025-09-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FfXG2GPt5F25tHRTPhAoZ4K.eval",
      "sourceId": "fXG2GPt5F25tHRTPhAoZ4K",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "qwen3.8-max_xhigh",
      "score": 0.458,
      "bestScore": 0.458,
      "standardError": 0.015763390640483554,
      "startedAt": "2026-08-27T19:30:55.000Z",
      "releaseDate": "2026-08-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fbfftd5EzUK9yAekafEofvK.eval",
      "sourceId": "bfftd5EzUK9yAekafEofvK",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gemini-3.5-flash_high",
      "score": 0.662,
      "bestScore": 0.662,
      "standardError": 0.014965960710224383,
      "startedAt": "2026-08-27T19:30:35.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGJ6YsSuzyRD9YXe6SBwvsu.eval",
      "sourceId": "GJ6YsSuzyRD9YXe6SBwvsu",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "deepseek-v4-pro_max",
      "score": 0.46993987975951906,
      "bestScore": 0.46993987975951906,
      "standardError": 0.015806515332355317,
      "startedAt": "2026-08-27T19:30:29.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FSq9xqkVTmFvELpQy9KH5TM.eval",
      "sourceId": "Sq9xqkVTmFvELpQy9KH5TM",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gemini-3.7-flash_high",
      "score": 0.692,
      "bestScore": 0.692,
      "standardError": 0.01460648312734278,
      "startedAt": "2026-08-27T19:30:25.000Z",
      "releaseDate": "2026-08-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F84ARuGyhQsjgh8umRmBvas.eval",
      "sourceId": "84ARuGyhQsjgh8umRmBvas",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "deepseek-v4-pro-0813_max",
      "score": 0.5290581162324649,
      "bestScore": 0.5290581162324649,
      "standardError": 0.015808394665706597,
      "startedAt": "2026-08-27T19:30:20.000Z",
      "releaseDate": "2026-08-13",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2xfFf6KZAp2gUWYhHGkFgs.eval",
      "sourceId": "2xfFf6KZAp2gUWYhHGkFgs",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "deepseek-v4-flash-0731_max",
      "score": 0.33633633633633636,
      "bestScore": 0.33633633633633636,
      "standardError": 0.01495531772489948,
      "startedAt": "2026-08-27T19:30:16.000Z",
      "releaseDate": "2026-07-31",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTqFmiGvqozt7TSZYmhgF8m.eval",
      "sourceId": "TqFmiGvqozt7TSZYmhgF8m",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gemini-3.6-flash_high",
      "score": 0.662,
      "bestScore": 0.662,
      "standardError": 0.014965960710224383,
      "startedAt": "2026-08-27T19:30:09.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FF83pa7Gfsb4UNVttTnS5Hk.eval",
      "sourceId": "F83pa7Gfsb4UNVttTnS5Hk",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gemma-4-31b-it",
      "score": 0.104,
      "bestScore": 0.104,
      "standardError": 0.009658016218524242,
      "startedAt": "2026-08-27T19:30:07.000Z",
      "releaseDate": "2026-04-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FD5cbXacPnJ8akq2jAtifEQ.eval",
      "sourceId": "D5cbXacPnJ8akq2jAtifEQ",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gemini-3-flash-preview_high",
      "score": 0.668,
      "bestScore": 0.668,
      "standardError": 0.014899597242811567,
      "startedAt": "2026-08-27T19:30:06.000Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7cc6cDCXjsPjtKt36bbeMR.eval",
      "sourceId": "7cc6cDCXjsPjtKt36bbeMR",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "grok-4.6_high",
      "score": 0.493,
      "bestScore": 0.493,
      "standardError": 0.01581774956184353,
      "startedAt": "2026-08-27T19:29:53.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTpHYtuPG6kYZFvMK7Z6Hqj.eval",
      "sourceId": "TpHYtuPG6kYZFvMK7Z6Hqj",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "grok-4.20-0309-reasoning",
      "score": 0.302,
      "bestScore": 0.302,
      "standardError": 0.014526080235459494,
      "startedAt": "2026-08-27T19:29:44.000Z",
      "releaseDate": "2026-02-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FGVm7khGvtdnQKGfkgAHD8q.eval",
      "sourceId": "GVm7khGvtdnQKGfkgAHD8q",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "grok-4.3_high",
      "score": 0.332,
      "bestScore": 0.332,
      "standardError": 0.014899597242811567,
      "startedAt": "2026-08-27T19:29:33.000Z",
      "releaseDate": "2026-04-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FY39ZRp553ULpUDBV2XrBbc.eval",
      "sourceId": "Y39ZRp553ULpUDBV2XrBbc",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "grok-4.5_high",
      "score": 0.483,
      "bestScore": 0.483,
      "standardError": 0.015810153729833274,
      "startedAt": "2026-08-27T19:29:30.000Z",
      "releaseDate": "2026-07-08",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F7HWW3sjRstKzb9JpcnvLaV.eval",
      "sourceId": "7HWW3sjRstKzb9JpcnvLaV",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "grok-4.6_xhigh",
      "score": 0.489,
      "bestScore": 0.489,
      "standardError": 0.015815471195292575,
      "startedAt": "2026-08-27T19:29:19.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FdKCWFaoGCWJTnD5CPWQGru.eval",
      "sourceId": "dKCWFaoGCWJTnD5CPWQGru",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "o4-mini-2025-04-16_low",
      "score": 0.196,
      "bestScore": 0.196,
      "standardError": 0.012559527926707347,
      "startedAt": "2026-08-27T19:29:17.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FDzky9brdTFbSoVd2zTKiUp.eval",
      "sourceId": "Dzky9brdTFbSoVd2zTKiUp",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.2-2025-12-11_low",
      "score": 0.328,
      "bestScore": 0.328,
      "standardError": 0.014853842487270312,
      "startedAt": "2026-08-27T19:29:12.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCyw3GA9a64AmT4pqLAhtTF.eval",
      "sourceId": "Cyw3GA9a64AmT4pqLAhtTF",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.2-2025-12-11_medium",
      "score": 0.327,
      "bestScore": 0.327,
      "standardError": 0.014842213153411162,
      "startedAt": "2026-08-27T19:29:07.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2Fesr5dcSuHiVoteo6vZfHjP.eval",
      "sourceId": "esr5dcSuHiVoteo6vZfHjP",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "o3-2025-04-16_high",
      "score": 0.494,
      "bestScore": 0.494,
      "standardError": 0.015818160898606833,
      "startedAt": "2026-08-27T19:28:51.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCxHSa2gY7ySiPPGTZEZaRt.eval",
      "sourceId": "CxHSa2gY7ySiPPGTZEZaRt",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.4-nano-2026-03-17_high",
      "score": 0.117,
      "bestScore": 0.117,
      "standardError": 0.010169287802713345,
      "startedAt": "2026-08-27T19:28:50.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F3QBsxKkbF2R2LEXTd6HDrJ.eval",
      "sourceId": "3QBsxKkbF2R2LEXTd6HDrJ",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "o4-mini-2025-04-16_high",
      "score": 0.188,
      "bestScore": 0.188,
      "standardError": 0.012361586015103758,
      "startedAt": "2026-08-27T19:28:50.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgaGdPvvfvAw7MjP4HaVQCT.eval",
      "sourceId": "gaGdPvvfvAw7MjP4HaVQCT",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.501,
      "bestScore": 0.501,
      "standardError": 0.015819268290576817,
      "startedAt": "2026-08-27T19:28:49.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FgTLurbyKU63i2opsGkJGwh.eval",
      "sourceId": "gTLurbyKU63i2opsGkJGwh",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.4-mini-2026-03-17_high",
      "score": 0.294,
      "bestScore": 0.294,
      "standardError": 0.014414290540008265,
      "startedAt": "2026-08-27T19:28:49.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZ3pjKgAULY9j9wmzQpdqZG.eval",
      "sourceId": "Z3pjKgAULY9j9wmzQpdqZG",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.2-2025-12-11_high",
      "score": 0.343,
      "bestScore": 0.343,
      "standardError": 0.015019206922357007,
      "startedAt": "2026-08-27T19:28:37.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXp4qkZ32zjDeSYii2eQpnm.eval",
      "sourceId": "Xp4qkZ32zjDeSYii2eQpnm",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.5_xhigh",
      "score": 0.63,
      "bestScore": 0.63,
      "standardError": 0.015275252316519465,
      "startedAt": "2026-08-27T19:28:25.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FTijCafTTJUFrJ6jPGvBbgq.eval",
      "sourceId": "TijCafTTJUFrJ6jPGvBbgq",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.4-pro-2026-03-05_xhigh",
      "score": 0.463,
      "bestScore": 0.463,
      "standardError": 0.015775927227262426,
      "startedAt": "2026-08-27T19:28:24.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FRfSJbuuuMkeVHQj7xrth4u.eval",
      "sourceId": "RfSJbuuuMkeVHQj7xrth4u",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.1-2025-11-13_high",
      "score": 0.48,
      "bestScore": 0.48,
      "standardError": 0.015806639423035177,
      "startedAt": "2026-08-27T19:28:24.000Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQCLnRy69NBH9ip3VYtQneh.eval",
      "sourceId": "QCLnRy69NBH9ip3VYtQneh",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-sonnet-4-5-20250929_59K",
      "score": 0.307,
      "bestScore": 0.307,
      "standardError": 0.014593284892852741,
      "startedAt": "2026-08-27T19:28:19.000Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FXQEGUQjpXupjdyKoGuSmum.eval",
      "sourceId": "XQEGUQjpXupjdyKoGuSmum",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.4-2026-03-05_xhigh",
      "score": 0.451,
      "bestScore": 0.451,
      "standardError": 0.015743152379585443,
      "startedAt": "2026-08-27T19:28:16.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FAzCkm282PLhDdLJ8DnAQQc.eval",
      "sourceId": "AzCkm282PLhDdLJ8DnAQQc",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-sonnet-5_xhigh",
      "score": 0.329,
      "bestScore": 0.329,
      "standardError": 0.014865395385928355,
      "startedAt": "2026-08-27T19:28:11.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FQw4tGw38pk4sdsEehhrLSR.eval",
      "sourceId": "Qw4tGw38pk4sdsEehhrLSR",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-opus-4-5-20251101_32K",
      "score": 0.457,
      "bestScore": 0.457,
      "standardError": 0.0157606915901365,
      "startedAt": "2026-08-27T19:28:02.000Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FCA5LW4phRnHSsZUGXEm8H6.eval",
      "sourceId": "CA5LW4phRnHSsZUGXEm8H6",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-haiku-4-5-20251001_32K",
      "score": 0.126,
      "bestScore": 0.126,
      "standardError": 0.010499249222407964,
      "startedAt": "2026-08-27T19:27:58.000Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FVhBM9HgJ2RuSVwXriRXMVJ.eval",
      "sourceId": "VhBM9HgJ2RuSVwXriRXMVJ",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.2-2025-12-11_xhigh",
      "score": 0.371,
      "bestScore": 0.371,
      "standardError": 0.015283736211823095,
      "startedAt": "2026-08-27T19:27:55.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FE3325ytkSNYP8ez53VA4fE.eval",
      "sourceId": "E3325ytkSNYP8ez53VA4fE",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-opus-4-6_max",
      "score": 0.47,
      "bestScore": 0.47,
      "standardError": 0.015790799515836722,
      "startedAt": "2026-08-27T19:27:45.000Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F85fhMxLaebWtLoi6oLpTbi.eval",
      "sourceId": "85fhMxLaebWtLoi6oLpTbi",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-sonnet-5_max",
      "score": 0.337,
      "bestScore": 0.337,
      "standardError": 0.014955087918653461,
      "startedAt": "2026-08-27T19:27:42.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FLPo74T9L3CawnKX5LLUrSZ.eval",
      "sourceId": "LPo74T9L3CawnKX5LLUrSZ",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-sonnet-4-6_max",
      "score": 0.328,
      "bestScore": 0.328,
      "standardError": 0.014853842487270312,
      "startedAt": "2026-08-27T19:27:37.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FZs5fCvNzaz8LhqgwDSAztU.eval",
      "sourceId": "Zs5fCvNzaz8LhqgwDSAztU",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-opus-4-7_xhigh",
      "score": 0.517,
      "bestScore": 0.517,
      "standardError": 0.015810153729833274,
      "startedAt": "2026-08-27T19:27:26.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2F2c4rvmf5N7QXTugrFLMZ83.eval",
      "sourceId": "2c4rvmf5N7QXTugrFLMZ83",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-opus-4-8_max",
      "score": 0.53,
      "bestScore": 0.53,
      "standardError": 0.015790799515836722,
      "startedAt": "2026-08-27T19:27:24.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": "https://logs.epoch.ai/inspect-viewer/36231d6d/viewer.html?log_file=https%3A%2F%2Flogs.epoch.ai%2Finspect_ai_logs%2FViC6pxQZUVS3E9M9iEubF7.eval",
      "sourceId": "ViC6pxQZUVS3E9M9iEubF7",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.6-sol_max",
      "score": 0.697,
      "bestScore": 0.697,
      "standardError": 0.014539683710535193,
      "startedAt": "2026-08-10T23:44:30.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "UnRUSuZzm2X27TUbpeMPva",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.6-terra_max",
      "score": 0.432,
      "bestScore": 0.432,
      "standardError": 0.015672320237336324,
      "startedAt": "2026-08-10T23:44:21.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "DienatPbgfgn3EzM7Gvzxr",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5.6-luna_max",
      "score": 0.41,
      "bestScore": 0.41,
      "standardError": 0.015560917136921657,
      "startedAt": "2026-08-10T23:44:20.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "CeDghdBk8CqtiFSbUzxNH5",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-opus-5_max",
      "score": 0.599,
      "bestScore": 0.599,
      "standardError": 0.01550610974549837,
      "startedAt": "2026-08-10T23:44:04.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Zftz2yk4X5NF28JQzjN3hZ",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-fable-5_xhigh",
      "score": 0.707,
      "bestScore": 0.707,
      "standardError": 0.014399942998441292,
      "startedAt": "2026-08-10T23:44:03.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "SEnvgtunRa5GwAtU2sHa87",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-sonnet-4-6_high",
      "score": 0.355,
      "bestScore": 0.355,
      "standardError": 0.015139491543780595,
      "startedAt": "2026-08-10T22:42:23.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LJyKf8Xy4dymxNHbGLwNdA",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-sonnet-4-5-20250929",
      "score": 0.237,
      "bestScore": 0.237,
      "standardError": 0.013454070462577935,
      "startedAt": "2026-08-10T22:42:10.000Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "MhqeuRVx4JcRfxPQDYPq7Z",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gemini-3.1-pro-preview_high",
      "score": 0.735,
      "bestScore": 0.735,
      "standardError": 0.013963164754810065,
      "startedAt": "2026-08-10T21:35:39.000Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "VGeF8Xac4WcSAGgSXVdEBi",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "kimi-k2.6",
      "score": 0.349,
      "bestScore": 0.349,
      "standardError": 0.015080663991563052,
      "startedAt": "2026-08-10T21:35:30.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "hHnQVatVTbQjM4sfrjw7GH",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5-mini-2025-08-07_high",
      "score": 0.216,
      "bestScore": 0.216,
      "standardError": 0.013019735539307761,
      "startedAt": "2026-08-10T21:35:29.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8d5RuFTTJKuJ5iJYmPZVJe",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "gpt-5-nano-2025-08-07_high",
      "score": 0.117,
      "bestScore": 0.117,
      "standardError": 0.010169287802713343,
      "startedAt": "2026-08-10T21:35:25.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "KgUCEfVCyRnY8B5kVChyiD",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "SimpleQA Verified",
      "model": "claude-haiku-4-5-20251001",
      "score": 0.132,
      "bestScore": 0.132,
      "standardError": 0.010709373963527967,
      "startedAt": "2026-08-10T21:26:22.000Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "J2kUEjR9r6Spd594qDNTTf",
      "sourceFile": "simpleqa_verified.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "mistral-large-4-0_high",
      "score": 0.33,
      "bestScore": 0.33,
      "standardError": 0.04725815626252608,
      "startedAt": "2026-10-15T22:52:59.000Z",
      "releaseDate": "2026-10-06",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "eNdDCt2u6A8mPisK56SApU",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-nano-2026-03-17_xhigh",
      "score": 0.45,
      "bestScore": 0.45,
      "standardError": 0.05,
      "startedAt": "2026-10-14T18:22:26.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "MeFyTsrHFWVrQJ8QHRjnRw",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.2-2025-12-11_xhigh",
      "score": 0.48,
      "bestScore": 0.48,
      "standardError": 0.05021167315686782,
      "startedAt": "2026-10-14T16:16:30.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "YGX23vJ7GJBaYUVLvHnXn5",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5-pro-2025-10-06_high",
      "score": 0.48,
      "bestScore": 0.48,
      "standardError": 0.05021167315686782,
      "startedAt": "2026-10-14T09:56:39.000Z",
      "releaseDate": "2025-10-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RwHkGk6Qhd4NVTtRLDwLmc",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.2-pro-2025-12-11_xhigh",
      "score": 0.6,
      "bestScore": 0.6,
      "standardError": 0.0492365963917331,
      "startedAt": "2026-10-14T09:56:34.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LgRDo2HcNECAGiD8UnHYjx",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-pro-2026-03-05_xhigh",
      "score": 0.59,
      "bestScore": 0.59,
      "standardError": 0.049431107042371025,
      "startedAt": "2026-10-14T09:56:24.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5yc9mNtsSiFQ5PducfGZGT",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.5-pro_xhigh",
      "score": 0.62,
      "bestScore": 0.62,
      "standardError": 0.04878317312145634,
      "startedAt": "2026-10-14T09:56:16.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5u4XrDnQSRnuXSK6oXKAth",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.6-sol_promax",
      "score": 0.68,
      "bestScore": 0.68,
      "standardError": 0.04688261722621507,
      "startedAt": "2026-10-14T09:56:09.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "WAavFdihYDHwjGsDNh9toF",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "glm-4.7",
      "score": 0.21,
      "bestScore": 0.21,
      "standardError": 0.040936018074033236,
      "startedAt": "2026-10-14T09:36:09.000Z",
      "releaseDate": "2025-12-22",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "9rDLegdh3zmBfFzTj3fkCc",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "glm-5.1",
      "score": 0.2,
      "bestScore": 0.2,
      "standardError": 0.04020151261036849,
      "startedAt": "2026-10-14T06:28:37.000Z",
      "releaseDate": "2026-04-07",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "R7XgJMi2R9sv7tZdx5SZTv",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "glm-5.2_max",
      "score": 0.22,
      "bestScore": 0.22,
      "standardError": 0.041633319989322654,
      "startedAt": "2026-10-14T05:10:27.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "Hui8iZgjwKPxz7DkRU8sJX",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "glm-5.2_low",
      "score": 0.18,
      "bestScore": 0.18,
      "standardError": 0.03861229196653691,
      "startedAt": "2026-10-14T04:18:44.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "6gYpimjqJm3CL6mvKEN9qs",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "glm-5.2_none",
      "score": 0.19,
      "bestScore": 0.19,
      "standardError": 0.039427724440366255,
      "startedAt": "2026-10-14T03:42:13.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "Xi45sLZGyn5sutf7nKRHTk",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "glm-5.3_max",
      "score": 0.37,
      "bestScore": 0.37,
      "standardError": 0.048523658709390974,
      "startedAt": "2026-10-13T21:22:18.000Z",
      "releaseDate": "2026-08-14",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "kG3hbd8qLEd6fLAJHyQTiN",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "kimi-k2.6",
      "score": 0.2,
      "bestScore": 0.2,
      "standardError": 0.04020151261036849,
      "startedAt": "2026-10-13T17:16:43.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "ehqiuEDETZTqJ8zt6Mc7Ja",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-4-argon_high",
      "score": 0.59,
      "bestScore": 0.59,
      "standardError": 0.04943110704237103,
      "startedAt": "2026-10-13T17:15:18.000Z",
      "releaseDate": "2026-09-30",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GJtYeC7ZV4fYgN3wu4rRvq",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "kimi-k2.7-code",
      "score": 0.14,
      "bestScore": 0.14,
      "standardError": 0.0348735088019777,
      "startedAt": "2026-10-13T17:13:43.000Z",
      "releaseDate": "2026-06-12",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "bSyTSKD8RiRNZCGNu6MnFp",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "kimi-k3_max",
      "score": 0.36,
      "bestScore": 0.36,
      "standardError": 0.048241815132442176,
      "startedAt": "2026-10-13T17:13:35.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "nzQV9YvictnDF2EwxqTJQs",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "kimi-k3_high",
      "score": 0.37,
      "bestScore": 0.37,
      "standardError": 0.048523658709390974,
      "startedAt": "2026-10-13T17:10:13.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "X2QNPHFXuWosA6AiuvVtDv",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "kimi-k3_low",
      "score": 0.18,
      "bestScore": 0.18,
      "standardError": 0.03861229196653691,
      "startedAt": "2026-10-13T17:06:54.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "VPSX7EnwWuBrCPnpYGMUEM",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "glm-5.3-flash_max",
      "score": 0.21,
      "bestScore": 0.21,
      "standardError": 0.040936018074033236,
      "startedAt": "2026-10-13T15:54:34.000Z",
      "releaseDate": "2026-08-20",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "fWNxmiSUC7BX8whqvzssY8",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-4o-mini-2024-07-18",
      "score": 0.03,
      "bestScore": 0.03,
      "standardError": 0.01714466079977653,
      "startedAt": "2026-10-13T14:12:48.000Z",
      "releaseDate": "2024-07-18",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "QGrgPmo24qW4h9SAvELP3Y",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o1-2024-12-17_high",
      "score": 0.1,
      "bestScore": 0.1,
      "standardError": 0.03015113445777636,
      "startedAt": "2026-10-13T14:12:19.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "UtRhB562A5f4SuFsg5MPrJ",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o1-2024-12-17_medium",
      "score": 0.08,
      "bestScore": 0.08,
      "standardError": 0.027265992434429075,
      "startedAt": "2026-10-13T14:12:16.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8hGkBXjvgrhMoviNRydRTy",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-4.1-2025-04-14",
      "score": 0.03,
      "bestScore": 0.03,
      "standardError": 0.01714466079977653,
      "startedAt": "2026-10-13T14:12:01.000Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "TzdbTbqAFB73Tfet8GFzQQ",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-4o-2024-08-06",
      "score": 0.06,
      "bestScore": 0.06,
      "standardError": 0.023868325657594204,
      "startedAt": "2026-10-13T14:12:01.000Z",
      "releaseDate": "2024-08-06",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Cjt9Xa7iPVBf9dN5cfJ88q",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o1-2024-12-17_low",
      "score": 0.09,
      "bestScore": 0.09,
      "standardError": 0.028762349126466143,
      "startedAt": "2026-10-13T14:12:00.000Z",
      "releaseDate": "2024-12-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GX3T3cWkC6n2Dc6ufeQN6y",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o3-mini-2025-01-31_medium",
      "score": 0.11,
      "bestScore": 0.11,
      "standardError": 0.031446603773522014,
      "startedAt": "2026-10-13T14:11:33.000Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "KX6LwuDrXwKvJ6RXod52j4",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-4.1-nano-2025-04-14",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2026-10-13T14:11:28.000Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "FLJ6bNrtaJQGCU2dP3CvWH",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o3-mini-2025-01-31_high",
      "score": 0.15,
      "bestScore": 0.15,
      "standardError": 0.03588702812826367,
      "startedAt": "2026-10-13T14:11:18.000Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "65oGRSHSEY6edS9mqbsCfE",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o3-mini-2025-01-31_low",
      "score": 0.06,
      "bestScore": 0.06,
      "standardError": 0.023868325657594204,
      "startedAt": "2026-10-13T14:11:15.000Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gE8uyjscAzcjpuZZNow3d8",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o4-mini-2025-04-16_high",
      "score": 0.13,
      "bestScore": 0.13,
      "standardError": 0.03379976689896311,
      "startedAt": "2026-10-13T14:10:35.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4MyKFuqZDpWQCFpKuHbMyD",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-4.1-mini-2025-04-14",
      "score": 0.01,
      "bestScore": 0.01,
      "standardError": 0.009999999999999998,
      "startedAt": "2026-10-13T14:10:21.000Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZyRri9YZ8CNz6CrUWXFjBA",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o4-mini-2025-04-16_medium",
      "score": 0.22,
      "bestScore": 0.22,
      "standardError": 0.041633319989322654,
      "startedAt": "2026-10-13T14:10:20.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dfonReEmxqpRWn5kxGSSES",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o4-mini-2025-04-16_low",
      "score": 0.15,
      "bestScore": 0.15,
      "standardError": 0.03588702812826367,
      "startedAt": "2026-10-13T14:10:10.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "EjErEBJsumLKEs2LByVSst",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5-mini-2025-08-07_low",
      "score": 0.16,
      "bestScore": 0.16,
      "standardError": 0.036845294917747066,
      "startedAt": "2026-10-13T14:09:58.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Wk25aTN6ZVjg5BwRUdcYKo",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o3-2025-04-16_low",
      "score": 0.36,
      "bestScore": 0.36,
      "standardError": 0.048241815132442176,
      "startedAt": "2026-10-13T14:09:51.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mjrcPbSQbMApmXPnqhoawd",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5-mini-2025-08-07_high",
      "score": 0.24,
      "bestScore": 0.24,
      "standardError": 0.04292346959909278,
      "startedAt": "2026-10-13T14:09:48.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5tdGvpe8Ahh8xsUnSwhdAF",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o3-2025-04-16_high",
      "score": 0.36,
      "bestScore": 0.36,
      "standardError": 0.048241815132442176,
      "startedAt": "2026-10-13T14:09:35.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RBFypLgCtFRyC4gsCewHfj",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "o3-2025-04-16_medium",
      "score": 0.37,
      "bestScore": 0.37,
      "standardError": 0.048523658709390974,
      "startedAt": "2026-10-13T14:09:29.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "MiKNm9AZG7MxP6hibnYN8a",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5-nano-2025-08-07_high",
      "score": 0.22,
      "bestScore": 0.22,
      "standardError": 0.041633319989322654,
      "startedAt": "2026-10-13T14:09:27.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gQHCk4faPpTNupNZ3xmr4c",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5-nano-2025-08-07_low",
      "score": 0.15,
      "bestScore": 0.15,
      "standardError": 0.03588702812826367,
      "startedAt": "2026-10-13T14:09:19.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "kdPFdnbdEGaSw4krPzeJtY",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5-nano-2025-08-07_minimal",
      "score": 0.02,
      "bestScore": 0.02,
      "standardError": 0.014070529413628966,
      "startedAt": "2026-10-13T14:08:37.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "BdrAtxC8wKnKiaEgedzdYY",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5-mini-2025-08-07_minimal",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2026-10-13T14:08:10.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ELe4PgXXEJQ8TDtUCQRKae",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.36,
      "bestScore": 0.36,
      "standardError": 0.04824181513244218,
      "startedAt": "2026-10-13T14:07:43.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "CxCwksDLsyrgTJ5ievyAoq",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5-2025-08-07_medium",
      "score": 0.32,
      "bestScore": 0.32,
      "standardError": 0.04688261722621506,
      "startedAt": "2026-10-13T14:07:34.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4QRM7D5HeCwCvWVZqhJKRS",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5-2025-08-07_low",
      "score": 0.26,
      "bestScore": 0.26,
      "standardError": 0.0440844002276808,
      "startedAt": "2026-10-13T14:07:24.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iygB8rn4Xogj5N5x9W7ifT",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5-2025-08-07_minimal",
      "score": 0.1,
      "bestScore": 0.1,
      "standardError": 0.03015113445777636,
      "startedAt": "2026-10-13T14:06:12.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gz54krKiBUPt4BfJASJZqi",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.1-2025-11-13_none",
      "score": 0.04,
      "bestScore": 0.04,
      "standardError": 0.019694638556693237,
      "startedAt": "2026-10-13T14:05:48.000Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5HoXvfpXgnMT3mZ2a8ezkC",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.2-2025-12-11_medium",
      "score": 0.36,
      "bestScore": 0.36,
      "standardError": 0.04824181513244218,
      "startedAt": "2026-10-13T14:05:35.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "WpdE2FAG4gvAbvgemVXgPA",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.1-2025-11-13_high",
      "score": 0.42,
      "bestScore": 0.42,
      "standardError": 0.04960449637488583,
      "startedAt": "2026-10-13T14:05:29.000Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "HKQxqBGokn83tHr5wGJiZX",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.1-2025-11-13_low",
      "score": 0.22,
      "bestScore": 0.22,
      "standardError": 0.041633319989322654,
      "startedAt": "2026-10-13T14:05:25.000Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LfDm9pJ9Uttcjcacy9DGv3",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.2-2025-12-11_high",
      "score": 0.45,
      "bestScore": 0.45,
      "standardError": 0.05,
      "startedAt": "2026-10-13T14:05:16.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Y9hzf2ioQTW9ezDvCGzTCu",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.2-2025-12-11_low",
      "score": 0.26,
      "bestScore": 0.26,
      "standardError": 0.04408440022768081,
      "startedAt": "2026-10-13T14:04:39.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gajcfzsiUXo8AyTudT22bP",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.2-2025-12-11_none",
      "score": 0.02,
      "bestScore": 0.02,
      "standardError": 0.014070529413628968,
      "startedAt": "2026-10-13T14:04:13.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "MuBMyrNUmTRebT8AA8AY8g",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-2026-03-05_xhigh",
      "score": 0.48,
      "bestScore": 0.48,
      "standardError": 0.05021167315686782,
      "startedAt": "2026-10-13T14:04:04.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "92ZBraB6bB98s8raSiPgUw",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-2026-03-05_high",
      "score": 0.42,
      "bestScore": 0.42,
      "standardError": 0.04960449637488583,
      "startedAt": "2026-10-13T14:04:01.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "QxrNbLxE4V9iHrUVmVPFHk",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-2026-03-05_medium",
      "score": 0.37,
      "bestScore": 0.37,
      "standardError": 0.048523658709390974,
      "startedAt": "2026-10-13T14:03:43.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "YA6uTqem8RwiNreRbaarAz",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-2026-03-05_low",
      "score": 0.19,
      "bestScore": 0.19,
      "standardError": 0.039427724440366255,
      "startedAt": "2026-10-13T14:03:36.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8fpUqaaFFUfWWNZskndSpp",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-2026-03-05_none",
      "score": 0.03,
      "bestScore": 0.03,
      "standardError": 0.017144660799776525,
      "startedAt": "2026-10-13T14:03:23.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "TfUnmjxgDbgUvDAo6JrD9a",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-nano-2026-03-17_high",
      "score": 0.36,
      "bestScore": 0.36,
      "standardError": 0.048241815132442176,
      "startedAt": "2026-10-13T14:02:58.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "N5YuEXtrXeXuhDN2qRssSK",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-nano-2026-03-17_low",
      "score": 0.2,
      "bestScore": 0.2,
      "standardError": 0.04020151261036849,
      "startedAt": "2026-10-13T14:02:49.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "isrAz3SUJUJHGUxsk3sG6V",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-nano-2026-03-17_none",
      "score": 0.02,
      "bestScore": 0.02,
      "standardError": 0.014070529413628968,
      "startedAt": "2026-10-13T14:02:42.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "at7F8ntqX3M6rXA38mKkWr",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-mini-2026-03-17_xhigh",
      "score": 0.17,
      "bestScore": 0.17,
      "standardError": 0.03775251680686369,
      "startedAt": "2026-10-13T14:02:26.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8Wf3UWZzc4nxwzzPv9ckvF",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-mini-2026-03-17_high",
      "score": 0.23,
      "bestScore": 0.23,
      "standardError": 0.04229525846816507,
      "startedAt": "2026-10-13T13:56:43.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "fzgzPmgay5wxMGHjevRjrm",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.5_xhigh",
      "score": 0.63,
      "bestScore": 0.63,
      "standardError": 0.048523658709390974,
      "startedAt": "2026-10-13T13:54:48.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "DFWQcrbMegFAcEWZKWoJxC",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.4-mini-2026-03-17_none",
      "score": 0.02,
      "bestScore": 0.02,
      "standardError": 0.014070529413628966,
      "startedAt": "2026-10-13T13:54:47.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Jpv6S8xWEHQQTzNHD35dcw",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.5_low",
      "score": 0.3,
      "bestScore": 0.3,
      "standardError": 0.04605661864718382,
      "startedAt": "2026-10-13T13:54:20.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4EayGByBV2xodNBkrZMcFb",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.5_none",
      "score": 0.09,
      "bestScore": 0.09,
      "standardError": 0.028762349126466143,
      "startedAt": "2026-10-13T13:53:55.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "QfS2PrjUFMNnceRZmCmewP",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.6-luna_max",
      "score": 0.43,
      "bestScore": 0.43,
      "standardError": 0.0497569851956243,
      "startedAt": "2026-10-13T13:53:45.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LUfGWqWQim7rdBPgSpbrXS",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.6-luna_low",
      "score": 0.17,
      "bestScore": 0.17,
      "standardError": 0.03775251680686369,
      "startedAt": "2026-10-13T13:53:37.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "FagAUZ7m5PJHseDATvp2vf",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.6-luna_none",
      "score": 0.03,
      "bestScore": 0.03,
      "standardError": 0.01714466079977653,
      "startedAt": "2026-10-13T13:53:29.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5Q7zWYmoRik8mhABL4wpVR",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.6-terra_max",
      "score": 0.47,
      "bestScore": 0.47,
      "standardError": 0.05016135580465918,
      "startedAt": "2026-10-13T13:53:13.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "J2ZgtxyaQZfyDMPTPxS8nm",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.6-terra_low",
      "score": 0.27,
      "bestScore": 0.27,
      "standardError": 0.044619604333847374,
      "startedAt": "2026-10-13T13:51:29.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bA5j9K2shzZVFGTMQ5g6EU",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.6-terra_none",
      "score": 0.05,
      "bestScore": 0.05,
      "standardError": 0.02190429135575903,
      "startedAt": "2026-10-13T13:50:17.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "HzNnNaRmyNExSVaMjmFCHe",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.6-sol_max",
      "score": 0.59,
      "bestScore": 0.59,
      "standardError": 0.049431107042371025,
      "startedAt": "2026-10-13T13:50:12.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "oWqwMmwEd7LU3eU5Ftak64",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.6-sol_low",
      "score": 0.31,
      "bestScore": 0.31,
      "standardError": 0.04648231987117317,
      "startedAt": "2026-10-13T13:49:48.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "V6aVGtn9gAmRep2oBGsqVN",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-5.6-sol_none",
      "score": 0.07,
      "bestScore": 0.07,
      "standardError": 0.02564323999762428,
      "startedAt": "2026-10-13T13:49:41.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "WvV6QREDDXeDEGdrKztDsy",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-sonnet-4-5-20250929_59K",
      "score": 0.1,
      "bestScore": 0.1,
      "standardError": 0.03015113445777636,
      "startedAt": "2026-10-13T13:46:15.000Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RuQX7KA7owzqzf8Le9veeM",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-3-opus-20240229",
      "score": 0.03,
      "bestScore": 0.03,
      "standardError": 0.01714466079977653,
      "startedAt": "2026-10-13T13:45:40.000Z",
      "releaseDate": "2024-02-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "WAbi8sZpyNAZoNN3Gmg22y",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-sonnet-4-5-20250929_32K",
      "score": 0.1,
      "bestScore": 0.1,
      "standardError": 0.03015113445777636,
      "startedAt": "2026-10-13T13:45:23.000Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gJizntrYwhnbpCviv9mNBN",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "MiniMax-M3_none",
      "score": 0.07,
      "bestScore": 0.07,
      "standardError": 0.02564323999762428,
      "startedAt": "2026-10-13T13:44:41.000Z",
      "releaseDate": "2026-06-01",
      "organization": "MiniMax",
      "country": "China",
      "logViewer": null,
      "sourceId": "Ws3g7xtto2fZ7AhuXJfecf",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-sonnet-4-6_max",
      "score": 0.23,
      "bestScore": 0.23,
      "standardError": 0.04229525846816507,
      "startedAt": "2026-10-13T13:44:15.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6P9AJQmBzaPmE7ossbqLjz",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-sonnet-4-5-20250929",
      "score": 0.06,
      "bestScore": 0.06,
      "standardError": 0.023868325657594204,
      "startedAt": "2026-10-13T13:43:54.000Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "VajBDnUCyX7uqAfMTbi63j",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-haiku-4-5-20251001_59K",
      "score": 0.08,
      "bestScore": 0.08,
      "standardError": 0.027265992434429072,
      "startedAt": "2026-10-13T13:43:41.000Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6bmTzBjpMySeHn65gc5Hu5",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-4-6_max",
      "score": 0.21,
      "bestScore": 0.21,
      "standardError": 0.040936018074033236,
      "startedAt": "2026-10-13T13:43:19.000Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "aKc2rzsA6YPYwCkE7nhvvB",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-haiku-4-5-20251001_32K",
      "score": 0.04,
      "bestScore": 0.04,
      "standardError": 0.019694638556693237,
      "startedAt": "2026-10-13T13:43:11.000Z",
      "releaseDate": "2025-10-15",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "BLgwhLGGofDBzxoc34YFHK",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-4-5-20251101_59K",
      "score": 0.14,
      "bestScore": 0.14,
      "standardError": 0.0348735088019777,
      "startedAt": "2026-10-13T13:42:53.000Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "DPuTDKvupUqKgkdVj7Awsp",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-4-5-20251101_32K",
      "score": 0.05,
      "bestScore": 0.05,
      "standardError": 0.02190429135575903,
      "startedAt": "2026-10-13T13:42:41.000Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "V7seJ9yeHu4LYjrkomhEQm",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-6-astra_max",
      "score": 0.74,
      "bestScore": 0.74,
      "standardError": 0.0440844002276808,
      "startedAt": "2026-10-13T13:42:38.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "KaEir82ukWPm2Fq4FanY2M",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-4-5-20251101",
      "score": 0.05,
      "bestScore": 0.05,
      "standardError": 0.021904291355759033,
      "startedAt": "2026-10-13T13:42:24.000Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4afZ2hjLiJnPy5pD7TfE3K",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3-flash-preview_high",
      "score": 0.39,
      "bestScore": 0.39,
      "standardError": 0.04902071300001973,
      "startedAt": "2026-10-13T13:42:10.000Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "UUNZJtgib9XbCjLEdhtGaT",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-6-sol_max",
      "score": 0.56,
      "bestScore": 0.56,
      "standardError": 0.049888765156985884,
      "startedAt": "2026-10-13T13:41:42.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "N9uFn7EZFsvHNcHEwEy3J6",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "qwen3.5-plus",
      "score": 0.08,
      "bestScore": 0.08,
      "standardError": 0.027265992434429072,
      "startedAt": "2026-10-13T13:39:53.000Z",
      "releaseDate": "2026-02-16",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "XEaDaqM3oaPcrtQ6XH5fGS",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "qwen3.5-flash",
      "score": 0.14,
      "bestScore": 0.14,
      "standardError": 0.034873508801977704,
      "startedAt": "2026-10-13T13:39:44.000Z",
      "releaseDate": "2026-02-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "fEgRyzyFYZraNnkkbRVSSg",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "qwen3.6-plus",
      "score": 0.14,
      "bestScore": 0.14,
      "standardError": 0.0348735088019777,
      "startedAt": "2026-10-13T13:38:57.000Z",
      "releaseDate": "2026-03-31",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "SP9UFPzxVnsbzs2kExP2bh",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "deepseek-v4-pro_max",
      "score": 0.19,
      "bestScore": 0.19,
      "standardError": 0.039427724440366255,
      "startedAt": "2026-10-13T13:38:19.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "DSyqeLiA9LLZF4iaxsTVHL",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "deepseek-v4-pro_high",
      "score": 0.15,
      "bestScore": 0.15,
      "standardError": 0.03588702812826367,
      "startedAt": "2026-10-13T13:38:10.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "duoDEpEXouGJfYrjd5DbvN",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-6-luna_max",
      "score": 0.36,
      "bestScore": 0.36,
      "standardError": 0.04824181513244218,
      "startedAt": "2026-10-13T13:37:06.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Jevnk9hi9oRBbGcFu2JxBC",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-sonnet-4-6_high",
      "score": 0.19,
      "bestScore": 0.19,
      "standardError": 0.039427724440366255,
      "startedAt": "2026-10-13T13:36:50.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Atc4F5iuMZu3vNBWVGCBnv",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-sonnet-4-6_medium",
      "score": 0.16,
      "bestScore": 0.16,
      "standardError": 0.03684529491774706,
      "startedAt": "2026-10-13T13:36:47.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "htvM9QKQJ8TfBxTzDxfZfp",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-4-7_max",
      "score": 0.41,
      "bestScore": 0.41,
      "standardError": 0.04943110704237103,
      "startedAt": "2026-10-13T13:36:23.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XXPqVNR7WCFFA2Rnzjcbk6",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-4-7_xhigh",
      "score": 0.36,
      "bestScore": 0.36,
      "standardError": 0.04824181513244218,
      "startedAt": "2026-10-13T13:35:58.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RHbNnWqK6x8f6pR2AZ8VAq",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-4-7_low",
      "score": 0.26,
      "bestScore": 0.26,
      "standardError": 0.0440844002276808,
      "startedAt": "2026-10-13T13:35:50.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "NyAvSf73nUe84HuY6Vpjkh",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-4-8_max",
      "score": 0.52,
      "bestScore": 0.52,
      "standardError": 0.05021167315686782,
      "startedAt": "2026-10-13T13:35:27.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PEZpuEuU65bu3ZtvjLytHq",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-4-8_low",
      "score": 0.29,
      "bestScore": 0.29,
      "standardError": 0.045604802157206865,
      "startedAt": "2026-10-13T13:35:16.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "icTHLT2S2egxYtgbjQHUng",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-4-8_none",
      "score": 0.19,
      "bestScore": 0.19,
      "standardError": 0.039427724440366255,
      "startedAt": "2026-10-13T13:34:59.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "jDWEviMweCdzGxLfz8sdBm",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "grok-4.20-0309-reasoning",
      "score": 0.29,
      "bestScore": 0.29,
      "standardError": 0.045604802157206865,
      "startedAt": "2026-10-13T13:34:38.000Z",
      "releaseDate": "2026-02-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ETNzxNbPKwj4qxkdo3VVgT",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-fable-5_max",
      "score": 0.47,
      "bestScore": 0.47,
      "standardError": 0.05016135580465918,
      "startedAt": "2026-10-13T13:34:37.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bxS76CJuVws8iwWU2Wr2PM",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "grok-4.3_high",
      "score": 0.18,
      "bestScore": 0.18,
      "standardError": 0.03861229196653691,
      "startedAt": "2026-10-13T13:34:31.000Z",
      "releaseDate": "2026-04-17",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "m4WQuAN88hKsqbT9zXbg7z",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.1-pro-preview_high",
      "score": 0.51,
      "bestScore": 0.51,
      "standardError": 0.05024183937956913,
      "startedAt": "2026-10-13T13:34:13.000Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "FTDFCpXzDTgfgWBNdHLHGc",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.5-flash_high",
      "score": 0.45,
      "bestScore": 0.45,
      "standardError": 0.05,
      "startedAt": "2026-10-13T13:33:53.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XUDdWKUZBhdWPPQBRgx8Pw",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.5-flash_low",
      "score": 0.42,
      "bestScore": 0.42,
      "standardError": 0.04960449637488583,
      "startedAt": "2026-10-13T13:33:40.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "68KxyCJbfMSq8JBL3LRJfr",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "deepseek-v4-pro_none",
      "score": 0.04,
      "bestScore": 0.04,
      "standardError": 0.019694638556693237,
      "startedAt": "2026-10-13T13:33:35.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "iKiaFJXjdHdUnbMXdKnq4Q",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.5-flash_minimal",
      "score": 0.27,
      "bestScore": 0.27,
      "standardError": 0.04461960433384737,
      "startedAt": "2026-10-13T13:33:29.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "H8CebpQyV2qEE8SQzjBrWe",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "qwen3.6-flash",
      "score": 0.12,
      "bestScore": 0.12,
      "standardError": 0.032659863237109045,
      "startedAt": "2026-10-13T13:33:28.000Z",
      "releaseDate": "2026-04-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "j99TDv5LAYUSvTUcR47T75",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "qwen3.7-max",
      "score": 0.26,
      "bestScore": 0.26,
      "standardError": 0.04408440022768081,
      "startedAt": "2026-10-13T13:33:00.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "cbQnfsMSA3aQaBtjm3aFx2",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "deepseek-v4-flash-0731_max",
      "score": 0.32,
      "bestScore": 0.32,
      "standardError": 0.04688261722621506,
      "startedAt": "2026-10-13T13:32:51.000Z",
      "releaseDate": "2026-07-31",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "Azz52eFkXwVJiRCGpaUBwB",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.5-flash-lite_high",
      "score": 0.16,
      "bestScore": 0.16,
      "standardError": 0.036845294917747066,
      "startedAt": "2026-10-13T13:26:52.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "JJyzHWKJUA3iTZDU8iYefz",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.5-flash-lite_low",
      "score": 0.21,
      "bestScore": 0.21,
      "standardError": 0.040936018074033236,
      "startedAt": "2026-10-13T13:25:11.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "d5QCJNq9uLtxjhiMhqXLmg",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-fable-5_high",
      "score": 0.45,
      "bestScore": 0.45,
      "standardError": 0.05,
      "startedAt": "2026-10-13T13:24:08.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9ZFhnyMNCKZ8XdR58WGD3D",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-fable-5_low",
      "score": 0.31,
      "bestScore": 0.31,
      "standardError": 0.04648231987117317,
      "startedAt": "2026-10-13T13:23:41.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "2jKf4iEfVQnHkgAwEtD9Nn",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-sonnet-5_max",
      "score": 0.45,
      "bestScore": 0.45,
      "standardError": 0.05,
      "startedAt": "2026-10-13T13:23:26.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PN9exWpeJGXJB8tZGX32e4",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-sonnet-5_xhigh",
      "score": 0.34,
      "bestScore": 0.34,
      "standardError": 0.04760952285695234,
      "startedAt": "2026-10-13T13:23:17.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "DFzqUCfxnQPUdK8b8D7aYu",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "MiniMax-M3",
      "score": 0.13,
      "bestScore": 0.13,
      "standardError": 0.03379976689896311,
      "startedAt": "2026-10-13T13:23:06.000Z",
      "releaseDate": "2026-06-01",
      "organization": "MiniMax",
      "country": "China",
      "logViewer": null,
      "sourceId": "CtmxkTPbFRofQWTkEw5aSe",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.5-flash-lite_minimal",
      "score": 0.07,
      "bestScore": 0.07,
      "standardError": 0.02564323999762428,
      "startedAt": "2026-10-13T13:19:48.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gzQH7StUXaxAtXgxkTgyxJ",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-5_max",
      "score": 0.43,
      "bestScore": 0.43,
      "standardError": 0.0497569851956243,
      "startedAt": "2026-10-13T13:19:00.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RLN98bmQSMnc5Zv39BF4M2",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.6-flash_high",
      "score": 0.4,
      "bestScore": 0.4,
      "standardError": 0.0492365963917331,
      "startedAt": "2026-10-13T13:14:52.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZDeGG5QYkdgVqzGPmmnbfx",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "grok-4.5_high",
      "score": 0.35,
      "bestScore": 0.35,
      "standardError": 0.047937248544110224,
      "startedAt": "2026-10-13T13:14:23.000Z",
      "releaseDate": "2026-07-08",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "jTuYE8VJTos73Kw77UFevq",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "grok-4.6_xhigh",
      "score": 0.33,
      "bestScore": 0.33,
      "standardError": 0.04725815626252608,
      "startedAt": "2026-10-13T13:14:03.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "NiuExVccWkWbWqi6ex6vvf",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "qwen3.8-max_xhigh",
      "score": 0.46,
      "bestScore": 0.46,
      "standardError": 0.05009082659620331,
      "startedAt": "2026-10-13T13:13:49.000Z",
      "releaseDate": "2026-08-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "bq5y6Tp83KLtxC58UDfpyN",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "deepseek-v4-pro-0813_max",
      "score": 0.46,
      "bestScore": 0.46,
      "standardError": 0.05009082659620331,
      "startedAt": "2026-10-13T13:13:34.000Z",
      "releaseDate": "2026-08-13",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "C56VHYLMeww2wXebeJh3gg",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "muse-spark-1.1_xhigh",
      "score": 0.34,
      "bestScore": 0.34,
      "standardError": 0.04760952285695234,
      "startedAt": "2026-10-13T13:13:33.000Z",
      "releaseDate": "2026-07-09",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4pbEvTxsbzNRCh9kU9Xhun",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-5_low",
      "score": 0.27,
      "bestScore": 0.27,
      "standardError": 0.044619604333847374,
      "startedAt": "2026-10-13T13:13:14.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XTW92PMsypptnTGBFuLCxe",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-fable-5-1_max",
      "score": 0.48,
      "bestScore": 0.48,
      "standardError": 0.05021167315686782,
      "startedAt": "2026-10-13T13:13:06.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "aLrHd8RCiNo3Yb4GqSawM7",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.6-flash_low",
      "score": 0.42,
      "bestScore": 0.42,
      "standardError": 0.04960449637488583,
      "startedAt": "2026-10-13T13:12:52.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hJ6XXLBm9yfXtSbbhWuwpw",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.6-flash_minimal",
      "score": 0.29,
      "bestScore": 0.29,
      "standardError": 0.04560480215720686,
      "startedAt": "2026-10-13T13:12:45.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "beSbPHiiWcwJBSzYRh7Kh4",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "muse-spark-1.2_xhigh",
      "score": 0.27,
      "bestScore": 0.27,
      "standardError": 0.044619604333847374,
      "startedAt": "2026-10-13T13:05:03.000Z",
      "releaseDate": "2026-08-05",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "FueK5bFAzGbkUao9JAibms",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-opus-5-5_max",
      "score": 0.67,
      "bestScore": 0.67,
      "standardError": 0.04725815626252609,
      "startedAt": "2026-10-13T13:01:31.000Z",
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "F2tuCm5VmNDAaYYh6qYdRy",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-sonnet-5-5_max",
      "score": 0.59,
      "bestScore": 0.59,
      "standardError": 0.04943110704237103,
      "startedAt": "2026-10-13T13:00:38.000Z",
      "releaseDate": "2026-09-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "kXU32N6tFkQLx5ei6nRJFS",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "muse-spark-1.3_max",
      "score": 0.37,
      "bestScore": 0.37,
      "standardError": 0.048523658709390974,
      "startedAt": "2026-10-13T13:00:28.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XkHyFXCWYDv2nAB5vmmWrU",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.7-flash_high",
      "score": 0.66,
      "bestScore": 0.66,
      "standardError": 0.04760952285695234,
      "startedAt": "2026-10-13T13:00:23.000Z",
      "releaseDate": "2026-08-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Xu34dKTAdpBWL6advf6DMg",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "qwen3.8-max-0902_xhigh",
      "score": 0.37,
      "bestScore": 0.37,
      "standardError": 0.048523658709390974,
      "startedAt": "2026-10-13T13:00:11.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "Ut4d4rvkUygdLTRomFb5rh",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "grok-4.6_high",
      "score": 0.33,
      "bestScore": 0.33,
      "standardError": 0.04725815626252608,
      "startedAt": "2026-10-13T13:00:07.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "YoAJyC3j8dZ6ZpDZ5rSDLw",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "Inkling_xhigh",
      "score": 0.21,
      "bestScore": 0.21,
      "standardError": 0.040936018074033236,
      "startedAt": "2026-10-13T12:49:23.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "UZ9LtVBCjHPMN9SKFUo7Wa",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gpt-6.1-sol_max",
      "score": 0.68,
      "bestScore": 0.68,
      "standardError": 0.04688261722621507,
      "startedAt": "2026-10-13T12:40:12.000Z",
      "releaseDate": "2026-09-29",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "KkGEgrjYDnofMoUJBCykcH",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-haiku-5-5_max",
      "score": 0.55,
      "bestScore": 0.55,
      "standardError": 0.049999999999999996,
      "startedAt": "2026-10-13T12:40:01.000Z",
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "KEyuvehfRGtcdgNfNavy62",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "deepseek-v4.1-flash_max",
      "score": 0.48,
      "bestScore": 0.48,
      "standardError": 0.05021167315686782,
      "startedAt": "2026-10-13T12:39:44.000Z",
      "releaseDate": "2026-09-09",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "ACpDTbhRPwce2BWQTWD8iC",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "muse-spark-1.3_xhigh",
      "score": 0.34,
      "bestScore": 0.34,
      "standardError": 0.04760952285695234,
      "startedAt": "2026-10-13T12:39:26.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Hi6xJczgqR4zfCXCXSF6rz",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "grok-4.7_xhigh",
      "score": 0.41,
      "bestScore": 0.41,
      "standardError": 0.04943110704237103,
      "startedAt": "2026-10-13T12:39:21.000Z",
      "releaseDate": "2026-09-21",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "L5Ck7pMy8bVmVMMLwxJN7f",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "gemini-3.8-flash_high",
      "score": 0.58,
      "bestScore": 0.58,
      "standardError": 0.04960449637488583,
      "startedAt": "2026-10-13T12:39:05.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Hpyy6CpmXmd5q7Zbdq63CR",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "Chess Puzzles",
      "model": "claude-haiku-5-5_low",
      "score": 0.24,
      "bestScore": 0.24,
      "standardError": 0.04292346959909278,
      "startedAt": "2026-10-13T07:50:03.000Z",
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "KSXLphjiiXtuGdqv9f7ENL",
      "sourceFile": "chess_puzzles.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "gemini-4-argon_high",
      "score": 0.6476190476190476,
      "bestScore": 0.6476190476190476,
      "standardError": 0.04902681019517619,
      "startedAt": "2026-10-13T17:16:18.000Z",
      "releaseDate": "2026-09-30",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mSJNKA7nCp2eZVqMo5nTUP",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "muse-wonder-1.0_max",
      "score": 0.3142857142857143,
      "bestScore": 0.3142857142857143,
      "standardError": 0.041513323271815936,
      "startedAt": "2026-10-12T21:51:35.000Z",
      "releaseDate": "2026-10-12",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "fMDaGtyboCNfTpP9aStwaU",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "claude-haiku-5-5_max",
      "score": 0.43809523809523804,
      "bestScore": 0.43809523809523804,
      "standardError": 0.017817416127494958,
      "startedAt": "2026-10-09T00:00:53.000Z",
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "FMiLkyqzaD56gA4uCCj9Qr",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "gpt-6.1-sol_max",
      "score": 0.5428571428571429,
      "bestScore": 0.5428571428571429,
      "standardError": 0.05753831415997416,
      "startedAt": "2026-09-29T18:13:25.000Z",
      "releaseDate": "2026-09-29",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LRUWNzFDcScwkTYuZYdeiV",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "claude-opus-5-5_max",
      "score": 0.7142857142857142,
      "bestScore": 0.7142857142857142,
      "standardError": 0.07063046178186345,
      "startedAt": "2026-09-22T21:38:58.000Z",
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iKWWUmcuzRjqisoshKVEpL",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "gpt-6-sol_max",
      "score": 0.5333333333333333,
      "bestScore": 0.5333333333333333,
      "standardError": 0.04616837959444122,
      "startedAt": "2026-09-22T21:35:18.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gM5SADcEixdr2P6PkseLeF",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "claude-fable-5-1_max",
      "score": 0.5714285714285714,
      "bestScore": 0.5714285714285714,
      "standardError": 0.06023386019368341,
      "startedAt": "2026-09-08T14:59:49.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8JukffGviNAYHXSdTJvswB",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "claude-opus-5_max",
      "score": 0.45714285714285713,
      "bestScore": 0.45714285714285713,
      "standardError": 0.03563483225498992,
      "startedAt": "2026-09-05T20:22:12.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mtHToEb2uzbaekANakjkJZ",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "gpt-5.6-sol_max",
      "score": 0.4476190476190475,
      "bestScore": 0.4476190476190475,
      "standardError": 0.041513323271815936,
      "startedAt": "2026-09-04T22:22:24.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "SjeXjsdda4NKRBTrPiJVM8",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "gpt-6-astra_max",
      "score": 0.7619047619047619,
      "bestScore": 0.7619047619047619,
      "standardError": 0.08384198505551908,
      "startedAt": "2026-08-30T15:38:49.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GhVYxWxttDxZ7akiJQPYsE",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "grok-4.6_xhigh",
      "score": 0.30476190476190473,
      "bestScore": 0.30476190476190473,
      "standardError": 0.016187363535215188,
      "startedAt": "2026-08-18T14:42:38.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "VvXTUMGEJiQqgvGMVmaL4m",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "claude-opus-4-8_max",
      "score": 0.2857142857142857,
      "bestScore": 0.2857142857142857,
      "standardError": 0.014197256999998662,
      "startedAt": "2026-08-07T15:44:07.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "3zda4kRmrZfxiafpk8BtMM",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "gpt-5.5_xhigh",
      "score": 0.3428571428571428,
      "bestScore": 0.3428571428571428,
      "standardError": 0.01982539047110602,
      "startedAt": "2026-07-27T13:22:32.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RESzhBbTKsxwmsbBNyXCEz",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "claude-fable-5_max",
      "score": 0.395238,
      "bestScore": 0.395238,
      "standardError": 0.020141,
      "startedAt": "2026-07-17T06:36:20.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "h6c9FVqkCj88ZD7ohCipuR",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "claude-opus-4-7_max",
      "score": 0.19047619047619047,
      "bestScore": 0.19047619047619047,
      "standardError": null,
      "startedAt": "2026-06-30T19:24:36.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Ad4geQVekrBqJn5Ka84awK",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "claude-opus-4-6_max",
      "score": 0.126984,
      "bestScore": 0.126984,
      "standardError": null,
      "startedAt": "2026-06-29T19:12:59.000Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "XSESb7YgzSTRENEhjHxuwc",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.126984,
      "bestScore": 0.126984,
      "standardError": null,
      "startedAt": "2026-06-29T19:12:20.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "VneAiVua3Fv2ySMKyyf4Qp",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "glm-5.2_max",
      "score": 0.09523809523809523,
      "bestScore": 0.09523809523809523,
      "standardError": null,
      "startedAt": "2026-06-29T13:38:16.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "U3Fqei2qExVeQDCizpSVfx",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "gpt-5.2-2025-12-11_xhigh",
      "score": 0.230159,
      "bestScore": 0.230159,
      "standardError": null,
      "startedAt": "2026-06-26T13:32:47.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "3542DFKGpjD6k8AZfGKD7W",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "qwen3.7-max",
      "score": 0.09523809523809523,
      "bestScore": 0.09523809523809523,
      "standardError": null,
      "startedAt": "2026-06-25T01:53:10.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "YJY5Dm2gEioHP2ExmTadqG",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "gpt-5.4-2026-03-05_xhigh",
      "score": 0.253968,
      "bestScore": 0.253968,
      "standardError": null,
      "startedAt": "2026-06-25T01:47:13.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RNXPxRZ4vMiwbspiqETy8B",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "gemini-3.5-flash_high",
      "score": 0.047619047619047616,
      "bestScore": 0.047619047619047616,
      "standardError": null,
      "startedAt": "2026-06-25T01:46:45.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "7QDCZBGyG6w6BxE3gw6Gbv",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "claude-opus-4-5-20251101_128K",
      "score": 0.14285714285714285,
      "bestScore": 0.14285714285714285,
      "standardError": null,
      "startedAt": "2026-06-25T01:45:41.000Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "eTTktoMp5sTthdSqVkLTST",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "kimi-k2.6",
      "score": 0.023809523809523808,
      "bestScore": 0.023809523809523808,
      "standardError": null,
      "startedAt": "2026-06-25T01:44:03.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "QwcSB4hQADcMNQg7EXmaGP",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "claude-sonnet-4-5-20250929",
      "score": 0.023809523809523808,
      "bestScore": 0.023809523809523808,
      "standardError": null,
      "startedAt": "2026-06-25T01:43:43.000Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "cUQEu2ysx4CcVJHi9KtzjQ",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "claude-opus-4-1-20250805",
      "score": 0.079365,
      "bestScore": 0.079365,
      "standardError": null,
      "startedAt": "2026-06-25T01:43:14.000Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "NMncsmkDVG6ZmNufPxNs2m",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "EBR-bench",
      "model": "gemini-3.1-pro-preview",
      "score": 0.14285714285714285,
      "bestScore": 0.14285714285714285,
      "standardError": null,
      "startedAt": "2026-06-25T01:42:49.000Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GYNraPqbBYhqqAyhYzJfQ7",
      "sourceFile": "ebr_bench.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "muse-wonder-1.0_max",
      "score": 0.51,
      "bestScore": 0.51,
      "standardError": 0.05024183937956913,
      "startedAt": "2026-10-14T16:17:18.000Z",
      "releaseDate": "2026-10-12",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "gmscEuzbtZeUeFUFZMZiAV",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "mistral-large-4-0_high",
      "score": 0.17,
      "bestScore": 0.17,
      "standardError": 0.03775251680686369,
      "startedAt": "2026-10-13T22:01:32.000Z",
      "releaseDate": "2026-10-06",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "LT9xLJnaU4UHsBQrvu3rbU",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-4-argon_high",
      "score": 0.64,
      "bestScore": 0.64,
      "standardError": 0.048241815132442176,
      "startedAt": "2026-10-13T17:14:57.000Z",
      "releaseDate": "2026-09-30",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9ipFgNR5MDw7ustUsuSviu",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "deepseek-v4.1-flash_max",
      "score": 0.43,
      "bestScore": 0.43,
      "standardError": 0.0497569851956243,
      "startedAt": "2026-10-08T19:02:32.000Z",
      "releaseDate": "2026-09-09",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "K3JABAsJCbXU5YrCPH2yVW",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-haiku-5-5_max",
      "score": 0.3,
      "bestScore": 0.3,
      "standardError": 0.04605661864718382,
      "startedAt": "2026-10-07T20:51:20.000Z",
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "CKm7NWXfSePfcUycRK87cB",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-6.1-sol_max",
      "score": 0.8,
      "bestScore": 0.8,
      "standardError": 0.04020151261036849,
      "startedAt": "2026-09-29T22:00:34.000Z",
      "releaseDate": "2026-09-29",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "fQzkNRV69sKaJX7o9KKneQ",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-5-5_max",
      "score": 0.71,
      "bestScore": 0.71,
      "standardError": 0.04560480215720687,
      "startedAt": "2026-09-29T17:34:54.000Z",
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "fREi6AAmnfqUWkDqUhLLtw",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-sonnet-5-5_max",
      "score": 0.65,
      "bestScore": 0.65,
      "standardError": 0.04793724854411022,
      "startedAt": "2026-09-29T15:15:14.000Z",
      "releaseDate": "2026-09-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "3PVdaSEYFJKgYUgHnF4DBu",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-6-luna_max",
      "score": 0.07,
      "bestScore": 0.07,
      "standardError": 0.02564323999762428,
      "startedAt": "2026-09-22T22:14:05.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bAERV8wdmoHMLz2K4bZCSK",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "grok-4.7_xhigh",
      "score": 0.29,
      "bestScore": 0.29,
      "standardError": 0.045604802157206865,
      "startedAt": "2026-09-22T21:49:22.000Z",
      "releaseDate": "2026-09-21",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Hk8bjr4u6KGqVE8SrPFKLq",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-6-sol_max",
      "score": 0.56,
      "bestScore": 0.56,
      "standardError": 0.049888765156985884,
      "startedAt": "2026-09-22T19:59:24.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "DnFTaE9qNsmBdYNqohhynY",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "muse-spark-1.3_max",
      "score": 0.25,
      "bestScore": 0.25,
      "standardError": 0.044640484731156636,
      "startedAt": "2026-09-18T19:12:25.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "6Rcc2q3USiUNB6U44bpvGH",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "muse-spark-1.3_xhigh",
      "score": 0.1414141414141414,
      "bestScore": 0.1414141414141414,
      "standardError": 0.03519858850692775,
      "startedAt": "2026-09-16T23:26:24.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "n6X597deBgzpbDSyxvhUAB",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.8-flash_high",
      "score": 0.47,
      "bestScore": 0.47,
      "standardError": 0.05016135580465918,
      "startedAt": "2026-09-02T17:15:36.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5yyuCCi2Dpfh6xLBMxqFyB",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-fable-5-1_max",
      "score": 0.58,
      "bestScore": 0.58,
      "standardError": 0.04960449637488583,
      "startedAt": "2026-09-01T18:45:06.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dBHxnnTzLGz3nHis5LET6t",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "glm-5.3_max",
      "score": 0.33,
      "bestScore": 0.33,
      "standardError": 0.04725815626252608,
      "startedAt": "2026-08-30T23:33:33.000Z",
      "releaseDate": "2026-08-14",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "UUq99Qhq3qizcoJo5vdhuK",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.4-2026-03-05_none",
      "score": 0.16,
      "bestScore": 0.16,
      "standardError": 0.03684529491774706,
      "startedAt": "2026-08-30T22:33:48.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GtW2cWyAXZPCszLi63e4yJ",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.6-luna_none",
      "score": 0.2,
      "bestScore": 0.2,
      "standardError": 0.04020151261036849,
      "startedAt": "2026-08-30T22:33:31.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PBrQpAgRCXeUSNveEn4aDE",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-6-astra_max",
      "score": 0.84,
      "bestScore": 0.84,
      "standardError": 0.036845294917747066,
      "startedAt": "2026-08-30T14:57:43.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "FA7pM9hKdkgTXmGHptcno7",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "kimi-k3_low",
      "score": 0.12,
      "bestScore": 0.12,
      "standardError": 0.032659863237109045,
      "startedAt": "2026-08-29T11:05:21.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "CAoLSNkdR27AmHb84fnr8B",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "kimi-k2.6_none",
      "score": 0.12,
      "bestScore": 0.12,
      "standardError": 0.03265986323710904,
      "startedAt": "2026-08-29T11:05:11.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "8iiRQiAoCepDELQfUv3VDq",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "glm-5.3-flash_max",
      "score": 0.08,
      "bestScore": 0.08,
      "standardError": 0.027265992434429072,
      "startedAt": "2026-08-28T14:38:27.000Z",
      "releaseDate": "2026-08-20",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "TMM3rPrAHcpBG5Pvw9kr4L",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.7-flash",
      "score": 0.15,
      "bestScore": 0.15,
      "standardError": 0.03588702812826367,
      "startedAt": "2026-08-28T07:12:24.000Z",
      "releaseDate": "2026-07-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "5e4Jp9dxtNisWx2wbRRZxM",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.6-terra_medium",
      "score": 0.1,
      "bestScore": 0.1,
      "standardError": 0.03015113445777636,
      "startedAt": "2026-08-28T07:12:23.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "WcCQQSHTFAhXk6qCWEeSLh",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.2-2025-12-11_medium",
      "score": 0.22,
      "bestScore": 0.22,
      "standardError": 0.041633319989322654,
      "startedAt": "2026-08-28T07:12:01.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "KHS8JvzY7n9EFgpcjEkj5v",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.5_low",
      "score": 0.28,
      "bestScore": 0.28,
      "standardError": 0.045126085985421296,
      "startedAt": "2026-08-28T07:11:53.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "QHfAMMDjjzQF6xitN2vBbA",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.7-plus",
      "score": 0.17,
      "bestScore": 0.17,
      "standardError": 0.03775251680686369,
      "startedAt": "2026-08-28T07:11:35.000Z",
      "releaseDate": "2026-06-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "DmptTSPXx5uM5TYZoV59JT",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.4-2026-03-05_medium",
      "score": 0.28,
      "bestScore": 0.28,
      "standardError": 0.04512608598542129,
      "startedAt": "2026-08-28T07:11:31.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ereDhnap4G3Q6ydsNQW97W",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-4-0613",
      "score": 0.12,
      "bestScore": 0.12,
      "standardError": 0.032659863237109045,
      "startedAt": "2026-08-28T07:08:14.000Z",
      "releaseDate": "2023-06-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "CvujZuqfj5r7qsa52Smyx9",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "o3-2025-04-16_high",
      "score": 0.29,
      "bestScore": 0.29,
      "standardError": 0.04560480215720687,
      "startedAt": "2026-08-27T19:27:05.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "FmLdgAHVM7MPn3JebBvxTb",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "o4-mini-2025-04-16_high",
      "score": 0.05,
      "bestScore": 0.05,
      "standardError": 0.02190429135575903,
      "startedAt": "2026-08-27T19:26:53.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "eJbFTXaxZjnW7HNR93u3WH",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "o3-2025-04-16_medium",
      "score": 0.23,
      "bestScore": 0.23,
      "standardError": 0.04229525846816507,
      "startedAt": "2026-08-27T19:22:47.000Z",
      "releaseDate": "2025-04-16",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "cxniYPW74ux4L6syqvPkaD",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.6-flash",
      "score": 0.12,
      "bestScore": 0.12,
      "standardError": 0.03265986323710904,
      "startedAt": "2026-08-27T19:18:34.000Z",
      "releaseDate": "2026-04-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "2y5jXmHGjyEyZvGjNDtRZe",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.4-2026-03-05_low",
      "score": 0.17,
      "bestScore": 0.17,
      "standardError": 0.03775251680686369,
      "startedAt": "2026-08-27T19:18:00.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "kWUUr3aHFZfKNARqAFxYZa",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.6-sol_low",
      "score": 0.26,
      "bestScore": 0.26,
      "standardError": 0.04408440022768081,
      "startedAt": "2026-08-27T19:17:45.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RMeUVa4tq3fURqCMtQWkfe",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.6-max-preview_none",
      "score": 0.19,
      "bestScore": 0.19,
      "standardError": 0.039427724440366255,
      "startedAt": "2026-08-27T19:13:41.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "mbySBWyYrKdtd7joxHyBdq",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "MiniMax-M3",
      "score": 0.07,
      "bestScore": 0.07,
      "standardError": 0.02564323999762428,
      "startedAt": "2026-08-27T19:13:33.000Z",
      "releaseDate": "2026-06-01",
      "organization": "MiniMax",
      "country": "China",
      "logViewer": null,
      "sourceId": "GGpgwGRFVATVJdQr8TgRc9",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.6-terra_low",
      "score": 0.19,
      "bestScore": 0.19,
      "standardError": 0.039427724440366255,
      "startedAt": "2026-08-27T19:13:32.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iE9NNTJWbGpRia2fUVGkPf",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.7-max_none",
      "score": 0.26,
      "bestScore": 0.26,
      "standardError": 0.0440844002276808,
      "startedAt": "2026-08-27T19:08:52.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "ENh7ZPQj7ihBePjsC8a7HV",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "o3-mini-2025-01-31_high",
      "score": 0.07,
      "bestScore": 0.07,
      "standardError": 0.02564323999762428,
      "startedAt": "2026-08-27T19:08:51.000Z",
      "releaseDate": "2025-01-31",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "JSmdEXGULNBJ7PfSoVoa7W",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.1-2025-11-13_medium",
      "score": 0.16,
      "bestScore": 0.16,
      "standardError": 0.03684529491774706,
      "startedAt": "2026-08-27T19:08:40.000Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8dvpS2RJKYLwpVh6M7AJkA",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5-2025-08-07_medium",
      "score": 0.15,
      "bestScore": 0.15,
      "standardError": 0.03588702812826367,
      "startedAt": "2026-08-27T19:08:19.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "3ZzqFaVH7RgvRyxCt6RseN",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.2-2025-12-11_low",
      "score": 0.1,
      "bestScore": 0.1,
      "standardError": 0.03015113445777636,
      "startedAt": "2026-08-27T19:03:44.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bLxTUZHmq3YkRejnh6baqZ",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5-2025-08-07_low",
      "score": 0.16,
      "bestScore": 0.16,
      "standardError": 0.03684529491774706,
      "startedAt": "2026-08-27T19:03:18.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "262FWwUwrjPpxVN8G8sDZT",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5-mini-2025-08-07_high",
      "score": 0.05,
      "bestScore": 0.05,
      "standardError": 0.021904291355759033,
      "startedAt": "2026-08-27T18:59:15.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "THodvrqzCUw7fjpxvKMMWT",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.6-luna_medium",
      "score": 0.12,
      "bestScore": 0.12,
      "standardError": 0.032659863237109045,
      "startedAt": "2026-08-27T18:48:42.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "BvShPsULJMkwxxf8vBSNfu",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.4-nano-2026-03-17_high",
      "score": 0.05,
      "bestScore": 0.05,
      "standardError": 0.021904291355759033,
      "startedAt": "2026-08-27T18:48:17.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "2e2NePZeU3iB4wvhwCaTci",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.6-27b_none",
      "score": 0.07,
      "bestScore": 0.07,
      "standardError": 0.02564323999762428,
      "startedAt": "2026-08-27T18:48:08.000Z",
      "releaseDate": "2026-04-22",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "dKZDn279PKuc2Kxm68D4r2",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.5-397b-a17b_none",
      "score": 0.18,
      "bestScore": 0.18,
      "standardError": 0.03861229196653691,
      "startedAt": "2026-08-27T18:47:58.000Z",
      "releaseDate": "2026-02-13",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "D7WDyCr24AVdzr4eKAreJu",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.1-2025-11-13_low",
      "score": 0.19,
      "bestScore": 0.19,
      "standardError": 0.039427724440366255,
      "startedAt": "2026-08-27T18:36:07.000Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iV2bSueom3C2SKw72ogiYr",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "glm-5.2_medium",
      "score": 0.15,
      "bestScore": 0.15,
      "standardError": 0.03588702812826367,
      "startedAt": "2026-08-27T18:36:04.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "NfN7VwMPjSVqViqrBQPWk2",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.5-122b-a10b_none",
      "score": 0.17,
      "bestScore": 0.17,
      "standardError": 0.03775251680686369,
      "startedAt": "2026-08-27T18:35:48.000Z",
      "releaseDate": "2026-02-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "X5FL4gkJQYjzJ7axz23rfd",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.4-mini-2026-03-17_medium",
      "score": 0.07,
      "bestScore": 0.07,
      "standardError": 0.02564323999762428,
      "startedAt": "2026-08-27T18:31:20.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4CzMJvWvuiLMCZhe2JGmSb",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.5-flash",
      "score": 0.2,
      "bestScore": 0.2,
      "standardError": 0.04020151261036849,
      "startedAt": "2026-08-27T18:27:45.000Z",
      "releaseDate": "2026-02-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "b9csPZ2c9CAsRXmzzfVoEW",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.6-plus_none",
      "score": 0.12,
      "bestScore": 0.12,
      "standardError": 0.03265986323710904,
      "startedAt": "2026-08-27T18:27:14.000Z",
      "releaseDate": "2026-03-31",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "jGgkDRHbodcaTzQ3Mp6SxS",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.5-plus_none",
      "score": 0.17,
      "bestScore": 0.17,
      "standardError": 0.03775251680686369,
      "startedAt": "2026-08-27T18:23:25.000Z",
      "releaseDate": "2026-02-16",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "XbtWdvRz65dbPPnHcSBvmF",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.6-35b-a3b_none",
      "score": 0.22,
      "bestScore": 0.22,
      "standardError": 0.041633319989322654,
      "startedAt": "2026-08-27T18:22:49.000Z",
      "releaseDate": "2026-04-14",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "Ua5RK6JRxdyrHRa3hVFQwq",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.4-mini-2026-03-17_low",
      "score": 0.08,
      "bestScore": 0.08,
      "standardError": 0.027265992434429072,
      "startedAt": "2026-08-27T18:22:45.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9zTyRhqbSKn7bbruvwDP4L",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.6-luna_low",
      "score": 0.17,
      "bestScore": 0.17,
      "standardError": 0.03775251680686369,
      "startedAt": "2026-08-27T18:18:35.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iTY94tKhVEcZsUZivMwLJa",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.7-plus_none",
      "score": 0.17,
      "bestScore": 0.17,
      "standardError": 0.03775251680686369,
      "startedAt": "2026-08-27T18:18:28.000Z",
      "releaseDate": "2026-06-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "a5qR5FNnftfR8pk9HJ3JUz",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.7-flash_none",
      "score": 0.14,
      "bestScore": 0.14,
      "standardError": 0.0348735088019777,
      "startedAt": "2026-08-27T18:18:13.000Z",
      "releaseDate": "2026-07-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "i7MVHsH5pikQKjYn6LBqHS",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.6-flash_none",
      "score": 0.18,
      "bestScore": 0.18,
      "standardError": 0.03861229196653691,
      "startedAt": "2026-08-27T18:14:08.000Z",
      "releaseDate": "2026-04-27",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "3QtkFWzSLzd4hgYGsNd5cL",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "glm-5.2_low",
      "score": 0.19,
      "bestScore": 0.19,
      "standardError": 0.039427724440366255,
      "startedAt": "2026-08-27T18:13:46.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "7DQNxGG5q6i7VtnzF7vNxA",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5-mini-2025-08-07_medium",
      "score": 0.04,
      "bestScore": 0.04,
      "standardError": 0.019694638556693237,
      "startedAt": "2026-08-27T18:13:19.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8RmBWoBLhvciTpz2oPSbDs",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "MiniMax-M3_none",
      "score": 0.08,
      "bestScore": 0.08,
      "standardError": 0.027265992434429075,
      "startedAt": "2026-08-27T18:09:34.000Z",
      "releaseDate": "2026-06-01",
      "organization": "MiniMax",
      "country": "China",
      "logViewer": null,
      "sourceId": "jWHk7BTYHQRXk2rXCbnpcS",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5-nano-2025-08-07_high",
      "score": 0.08,
      "bestScore": 0.08,
      "standardError": 0.027265992434429075,
      "startedAt": "2026-08-27T18:09:14.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "aW4ppDktWqoy5qVWMUws6z",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.4-nano-2026-03-17_medium",
      "score": 0.06,
      "bestScore": 0.06,
      "standardError": 0.0238683256575942,
      "startedAt": "2026-08-27T18:08:44.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "hRMMiGMPvHSGh3ygSjgrs2",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.6-sol_none",
      "score": 0.33,
      "bestScore": 0.33,
      "standardError": 0.04725815626252608,
      "startedAt": "2026-08-27T18:07:54.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4uxDyWmq5f5azVYfWBGUHh",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.5_none",
      "score": 0.18,
      "bestScore": 0.18,
      "standardError": 0.03861229196653691,
      "startedAt": "2026-08-27T18:07:51.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "MHAMdn52f7wLMZ6MSCs5E2",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3-max-2025-09-23",
      "score": 0.05,
      "bestScore": 0.05,
      "standardError": 0.021904291355759033,
      "startedAt": "2026-08-27T18:07:41.000Z",
      "releaseDate": "2025-09-24",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "SfzmmHYZjuvWnvoYWit58g",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-oss-120b_high",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2026-08-27T18:02:21.000Z",
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GbQrTQyihBUsb9sxbqT8pT",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.2-2025-12-11_none",
      "score": 0.14,
      "bestScore": 0.14,
      "standardError": 0.0348735088019777,
      "startedAt": "2026-08-27T18:02:02.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "eSfCVMMp2HXYqvpVpyNjjr",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "glm-5.2_minimal",
      "score": 0.18,
      "bestScore": 0.18,
      "standardError": 0.03861229196653691,
      "startedAt": "2026-08-27T18:01:49.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "aekViWbdqVVeXpEG5hrSbQ",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.5-flash_none",
      "score": 0.16,
      "bestScore": 0.16,
      "standardError": 0.036845294917747066,
      "startedAt": "2026-08-27T18:01:12.000Z",
      "releaseDate": "2026-02-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "DjFkVt62QwqQ4xXwzYR5Ng",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "deepseek-v4-pro_none",
      "score": 0.17,
      "bestScore": 0.17,
      "standardError": 0.03775251680686369,
      "startedAt": "2026-08-27T18:01:08.000Z",
      "releaseDate": "2026-04-24",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "BmvJaHMCnZzb24ErpAsozU",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.4-nano-2026-03-17_low",
      "score": 0.03,
      "bestScore": 0.03,
      "standardError": 0.01714466079977653,
      "startedAt": "2026-08-27T18:01:04.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "aAAtrj4DqLUuEY6P7W5yLe",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "nemotron-3-ultra",
      "score": 0.2,
      "bestScore": 0.2,
      "standardError": 0.04020151261036849,
      "startedAt": "2026-08-27T18:00:58.000Z",
      "releaseDate": "2026-06-04",
      "organization": "Nvidia",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "U8H8juE3Ko83jsR6TRntbh",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.6-terra_none",
      "score": 0.14,
      "bestScore": 0.14,
      "standardError": 0.0348735088019777,
      "startedAt": "2026-08-27T18:00:41.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "8grv57UmR89LXkEYhyssr9",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5-nano-2025-08-07_medium",
      "score": 0.09,
      "bestScore": 0.09,
      "standardError": 0.028762349126466143,
      "startedAt": "2026-08-27T18:00:36.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZwQdGcba6Hez4ztMzmZP9j",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5-2025-08-07_minimal",
      "score": 0.14,
      "bestScore": 0.14,
      "standardError": 0.0348735088019777,
      "startedAt": "2026-08-27T17:58:55.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "G3tHLT8qE9ncseZazqTXRX",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.1-2025-11-13_none",
      "score": 0.15,
      "bestScore": 0.15,
      "standardError": 0.03588702812826368,
      "startedAt": "2026-08-27T17:58:50.000Z",
      "releaseDate": "2025-11-13",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "2a2uhidAfksbnAJC8yUrZK",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-4.1-mini-2025-04-14",
      "score": 0.07,
      "bestScore": 0.07,
      "standardError": 0.02564323999762428,
      "startedAt": "2026-08-27T17:58:41.000Z",
      "releaseDate": "2025-04-14",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "giKk2ut3JVFVQUvbykxvSu",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5-mini-2025-08-07_minimal",
      "score": 0.1,
      "bestScore": 0.1,
      "standardError": 0.03015113445777636,
      "startedAt": "2026-08-27T17:58:37.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "kW6hSbkCwQczqbazduGXPR",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "glm-5.2_none",
      "score": 0.18,
      "bestScore": 0.18,
      "standardError": 0.03861229196653691,
      "startedAt": "2026-08-27T17:58:06.000Z",
      "releaseDate": "2026-06-16",
      "organization": "Z.ai (Zhipu AI)",
      "country": "China",
      "logViewer": null,
      "sourceId": "BLjXudxuEyLbZiKBSZz326",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "Qwen3-235B-A22B-Thinking-2507",
      "score": 0.09,
      "bestScore": 0.09,
      "standardError": 0.028762349126466143,
      "startedAt": "2026-08-27T17:58:05.000Z",
      "releaseDate": "2025-07-25",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "BP3KYD8rYyFg3LviNGtjLL",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-3.5-turbo-0125",
      "score": 0.03,
      "bestScore": 0.03,
      "standardError": 0.01714466079977653,
      "startedAt": "2026-08-27T17:57:53.000Z",
      "releaseDate": "2024-01-25",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "UfmijCsx59sZ8xpcbcWDms",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.4-mini-2026-03-17_none",
      "score": 0.11,
      "bestScore": 0.11,
      "standardError": 0.03144660377352201,
      "startedAt": "2026-08-27T17:57:44.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "NEysw7ag6KJxUahem2PHQv",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-oss-120b_medium",
      "score": 0.02,
      "bestScore": 0.02,
      "standardError": 0.014070529413628966,
      "startedAt": "2026-08-27T17:57:43.000Z",
      "releaseDate": "2025-08-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "58MQQiJ7vvgvP9e8LcG6eV",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5-nano-2025-08-07_low",
      "score": 0.09,
      "bestScore": 0.09,
      "standardError": 0.028762349126466143,
      "startedAt": "2026-08-27T17:57:33.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "PLbKjMLohFB9epj3qHPcuS",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5-nano-2025-08-07_minimal",
      "score": 0.05,
      "bestScore": 0.05,
      "standardError": 0.021904291355759033,
      "startedAt": "2026-08-27T17:56:54.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "RVRYWkwXwrvbCyZW8AhqEm",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.4-nano-2026-03-17_none",
      "score": 0.09,
      "bestScore": 0.09,
      "standardError": 0.028762349126466143,
      "startedAt": "2026-08-27T17:56:52.000Z",
      "releaseDate": "2026-03-17",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "fxNDgLaScEoHo2canW89yv",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-4o-mini-2024-07-18",
      "score": 0.12,
      "bestScore": 0.12,
      "standardError": 0.03265986323710904,
      "startedAt": "2026-08-27T17:56:46.000Z",
      "releaseDate": "2024-07-18",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mbviVfQ7Bh7wHSCDDmrt7u",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "deepseek-v4-pro-0813_max",
      "score": 0.43,
      "bestScore": 0.43,
      "standardError": 0.0497569851956243,
      "startedAt": "2026-08-19T18:14:49.000Z",
      "releaseDate": "2026-08-13",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "c2KVkMCWnpo6BbnMWgdaDH",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "inkling-small_xhigh",
      "score": 0.06,
      "bestScore": 0.06,
      "standardError": 0.023868325657594204,
      "startedAt": "2026-08-15T01:02:40.000Z",
      "releaseDate": "2026-07-15",
      "organization": "Thinking Machines",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "nPFA9ufjfVqg3YNmDRqWeC",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.7-flash_high",
      "score": 0.37,
      "bestScore": 0.37,
      "standardError": 0.048523658709390974,
      "startedAt": "2026-08-14T21:35:50.000Z",
      "releaseDate": "2026-08-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZavSMMq9jk5s9WSbARZr6X",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "grok-4.6_xhigh",
      "score": 0.34,
      "bestScore": 0.34,
      "standardError": 0.04760952285695234,
      "startedAt": "2026-08-14T20:27:46.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9cXtChGrowgGDwa9AcAecB",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.2-2025-12-11_high",
      "score": 0.23,
      "bestScore": 0.23,
      "standardError": 0.04229525846816508,
      "startedAt": "2026-08-06T14:07:21.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "AsFSc8AQu4F46Tzp4xVwYu",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-sonnet-4-6",
      "score": 0.14,
      "bestScore": 0.14,
      "standardError": 0.0348735088019777,
      "startedAt": "2026-08-06T00:18:43.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "VjfWxdhdNerrh4huSHT8PT",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-sonnet-4-6_low",
      "score": 0.16,
      "bestScore": 0.16,
      "standardError": 0.03684529491774706,
      "startedAt": "2026-08-06T00:18:14.000Z",
      "releaseDate": "2026-02-17",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "YdyvoMQ4mA5V8Ymg6Tksxu",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-5",
      "score": 0.37,
      "bestScore": 0.37,
      "standardError": 0.048523658709390974,
      "startedAt": "2026-08-06T00:14:11.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bWnggCv6to2QKSjngrtuos",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-4-7",
      "score": 0.13,
      "bestScore": 0.13,
      "standardError": 0.033799766898963114,
      "startedAt": "2026-08-06T00:13:47.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "LsnPrukDADRTP8fsn8zD9S",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-4-6",
      "score": 0.15,
      "bestScore": 0.15,
      "standardError": 0.03588702812826367,
      "startedAt": "2026-08-06T00:13:31.000Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "BbM4xbJcS5sjrzZABGg8K7",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-4-6_low",
      "score": 0.07,
      "bestScore": 0.07,
      "standardError": 0.02564323999762428,
      "startedAt": "2026-08-06T00:13:22.000Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "JarZHkwUMZ6HuA4ArMNT45",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.1-pro-preview_medium",
      "score": 0.32,
      "bestScore": 0.32,
      "standardError": 0.04688261722621506,
      "startedAt": "2026-08-06T00:12:06.000Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "WueZbBM3NwMUZPveuagrZo",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.1-pro-preview_low",
      "score": 0.29,
      "bestScore": 0.29,
      "standardError": 0.04560480215720687,
      "startedAt": "2026-08-06T00:11:58.000Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4a3M8rRWUoPphCSr2dguXH",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-sonnet-5",
      "score": 0.16,
      "bestScore": 0.16,
      "standardError": 0.036845294917747066,
      "startedAt": "2026-08-06T00:09:48.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "AZHoMLTNQEcKLwtv7QzBLZ",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.5-flash_low",
      "score": 0.28,
      "bestScore": 0.28,
      "standardError": 0.045126085985421296,
      "startedAt": "2026-08-05T23:51:58.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "nrh3xa3Bn765Q6DXWoVfbS",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.6-flash_minimal",
      "score": 0.25,
      "bestScore": 0.25,
      "standardError": 0.04351941398892446,
      "startedAt": "2026-08-05T23:51:57.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "3L22tfAid5ZM42p92tqC73",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.6-flash_low",
      "score": 0.23,
      "bestScore": 0.23,
      "standardError": 0.04229525846816508,
      "startedAt": "2026-08-05T23:51:40.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "M8sY8vHnvSLxQUg3HTMTCp",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3-flash-preview_minimal",
      "score": 0.25,
      "bestScore": 0.25,
      "standardError": 0.04351941398892446,
      "startedAt": "2026-08-05T23:47:51.000Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "njpkLAWZTvept6tnkraYrR",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3-flash-preview_high",
      "score": 0.2,
      "bestScore": 0.2,
      "standardError": 0.04020151261036849,
      "startedAt": "2026-08-05T23:47:26.000Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "MvdunHkVURna3JuyDjhzEc",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3-flash-preview_low",
      "score": 0.26,
      "bestScore": 0.26,
      "standardError": 0.0440844002276808,
      "startedAt": "2026-08-05T23:47:24.000Z",
      "releaseDate": "2025-12-17",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "ZQMcTQjHcDSevCN2dYYHKY",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.5-flash-lite_minimal",
      "score": 0.12,
      "bestScore": 0.12,
      "standardError": 0.03265986323710904,
      "startedAt": "2026-08-05T23:43:12.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "fJKhFRm5dn9wzbXojee7uj",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.5-flash-lite_low",
      "score": 0.19,
      "bestScore": 0.19,
      "standardError": 0.039427724440366255,
      "startedAt": "2026-08-05T23:42:56.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Mg2yb9Tp3yc2H3ApJX6JBK",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "deepseek-v4-flash-0731_max",
      "score": 0.34,
      "bestScore": 0.34,
      "standardError": 0.04760952285695233,
      "startedAt": "2026-08-05T23:40:26.000Z",
      "releaseDate": "2026-07-31",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "4S4GXbkRhLz8wWP97jYG57",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.5-plus",
      "score": 0.16,
      "bestScore": 0.16,
      "standardError": 0.036845294917747066,
      "startedAt": "2026-08-05T23:33:02.000Z",
      "releaseDate": "2026-02-16",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "kCQr6ARwJyUfkwWqKAfuDX",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5-2025-08-07_high",
      "score": 0.23,
      "bestScore": 0.23,
      "standardError": 0.04229525846816508,
      "startedAt": "2026-08-05T19:32:49.000Z",
      "releaseDate": "2025-08-07",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "NKdPCHXg9sphwKW2UEnmkD",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.6-flash_high",
      "score": 0.3,
      "bestScore": 0.3,
      "standardError": 0.04605661864718382,
      "startedAt": "2026-08-05T01:49:56.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "YBqiCXPjMUSAgAjDUoz5sy",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.8-max_xhigh",
      "score": 0.38,
      "bestScore": 0.38,
      "standardError": 0.048783173121456344,
      "startedAt": "2026-08-05T01:42:38.000Z",
      "releaseDate": "2026-08-02",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "c2YLifDU4zYCX5Jbi9HLL7",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-fable-5_max",
      "score": 0.52,
      "bestScore": 0.52,
      "standardError": 0.05021167315686782,
      "startedAt": "2026-07-31T19:05:04.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "X29V6KG3aVe5Nv5r3tsLHX",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "kimi-k3_max",
      "score": 0.26,
      "bestScore": 0.26,
      "standardError": 0.0440844002276808,
      "startedAt": "2026-07-29T03:26:08.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "eCq63CuekohAswYftWXJVj",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-sonnet-5_max",
      "score": 0.35,
      "bestScore": 0.35,
      "standardError": 0.04793724854411023,
      "startedAt": "2026-07-28T21:09:50.000Z",
      "releaseDate": "2026-06-30",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "YdcyYXLzqfPUSSgGv4MTED",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.6-luna_max",
      "score": 0.21,
      "bestScore": 0.21,
      "standardError": 0.040936018074033236,
      "startedAt": "2026-07-28T20:59:07.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Epg7eNLwwGPXtX3sSAwPf8",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.6-terra_max",
      "score": 0.35,
      "bestScore": 0.35,
      "standardError": 0.047937248544110224,
      "startedAt": "2026-07-28T20:58:46.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "MFTd4BUbtzbfcCky6d79d4",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.6-sol_max",
      "score": 0.58,
      "bestScore": 0.58,
      "standardError": 0.04960449637488583,
      "startedAt": "2026-07-28T20:58:29.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5kMRXeasRvtghXpDotCdgV",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "qwen3.7-max",
      "score": 0.32,
      "bestScore": 0.32,
      "standardError": 0.04688261722621506,
      "startedAt": "2026-07-28T19:23:45.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "f7dotY9NheKNKK2zz6bfKZ",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.5_high",
      "score": 0.52,
      "bestScore": 0.52,
      "standardError": 0.05021167315686782,
      "startedAt": "2026-07-27T19:10:39.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Knij32Jp6bQJ2V9NHpg5VS",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.1-pro-preview_high",
      "score": 0.34,
      "bestScore": 0.34,
      "standardError": 0.04760952285695234,
      "startedAt": "2026-07-27T18:29:16.000Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "g4bbCaiFDqw2YYNDAcmn2a",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gemini-3.5-flash_high",
      "score": 0.32,
      "bestScore": 0.32,
      "standardError": 0.04688261722621506,
      "startedAt": "2026-07-27T18:28:52.000Z",
      "releaseDate": "2026-05-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GQKsxFbdBWB546zE8ci49f",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-4-8_xhigh",
      "score": 0.31,
      "bestScore": 0.31,
      "standardError": 0.04648231987117317,
      "startedAt": "2026-07-26T17:43:35.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "3ZnPtGRJkvqHk8wvWWRgCe",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-sonnet-4-5-20250929_48K",
      "score": 0.17,
      "bestScore": 0.17,
      "standardError": 0.03775251680686369,
      "startedAt": "2026-07-25T07:06:00.000Z",
      "releaseDate": "2025-09-29",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4oSbA2RKFsfuAfVayAMAww",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-4-1-20250805_24K",
      "score": 0.21,
      "bestScore": 0.21,
      "standardError": 0.040936018074033236,
      "startedAt": "2026-07-25T07:05:43.000Z",
      "releaseDate": "2025-08-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "2AbeCjy9w4KJWKhbKjtFTJ",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-4-5-20251101_48K",
      "score": 0.22,
      "bestScore": 0.22,
      "standardError": 0.04163331998932266,
      "startedAt": "2026-07-25T07:05:27.000Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "k4zGHaXESxFRTRMkyDLZrz",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-4-8_max",
      "score": 0.36,
      "bestScore": 0.36,
      "standardError": 0.04824181513244218,
      "startedAt": "2026-07-25T04:37:20.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Q3WpHdurYNvawj6ycXyj6e",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-5_max",
      "score": 0.59,
      "bestScore": 0.59,
      "standardError": 0.04943110704237103,
      "startedAt": "2026-07-25T04:37:04.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "FfAvNS5H9bMDSBF4JNNiDz",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-4-6_max",
      "score": 0.25,
      "bestScore": 0.25,
      "standardError": 0.04351941398892446,
      "startedAt": "2026-07-25T01:44:09.000Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "64rokFyL2L8Vd4SZVdayCy",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "claude-opus-4-7_max",
      "score": 0.28,
      "bestScore": 0.28,
      "standardError": 0.04512608598542129,
      "startedAt": "2026-07-25T01:43:24.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "VFYhE54autgdYcMJNXicwU",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.4-2026-03-05_xhigh",
      "score": 0.37,
      "bestScore": 0.37,
      "standardError": 0.048523658709390974,
      "startedAt": "2026-07-24T18:20:01.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "T6DyawEuhVmDAaDTzDRNNx",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "gpt-5.5_xhigh",
      "score": 0.56,
      "bestScore": 0.56,
      "standardError": 0.049888765156985884,
      "startedAt": "2026-07-24T18:19:51.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "EatRQnSaUpUwKzdtjYnQHo",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "Mystery Game Puzzles",
      "model": "kimi-k2.6",
      "score": 0.18,
      "bestScore": 0.18,
      "standardError": 0.03861229196653691,
      "startedAt": "2026-07-17T23:54:57.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "RA5MRbmWJpFPR29BQxfvZt",
      "sourceFile": "mystery_game_puzzles.csv"
    },
    {
      "benchmark": "MirrorCode",
      "model": "claude-opus-5-5_max",
      "score": 0.774,
      "bestScore": 0.774,
      "standardError": 0.0831,
      "startedAt": "2026-09-22T16:42:00.000Z",
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mirrorcode-ml-2l-claude-opus-5-5",
      "sourceFile": "mirrorcode.csv"
    },
    {
      "benchmark": "MirrorCode",
      "model": "claude-fable-5-1_high",
      "score": 0.733,
      "bestScore": 0.733,
      "standardError": 0.0917,
      "startedAt": "2026-09-10T13:19:30.449Z",
      "releaseDate": "2026-09-01",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mirrorcode-ml-2l-claude-fable-5-1",
      "sourceFile": "mirrorcode.csv"
    },
    {
      "benchmark": "MirrorCode",
      "model": "gpt-6-astra_high",
      "score": 0.467,
      "bestScore": 0.467,
      "standardError": 0.1186,
      "startedAt": "2026-08-30T16:51:09.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mirrorcode-ml-2l-gpt-6-astra",
      "sourceFile": "mirrorcode.csv"
    },
    {
      "benchmark": "MirrorCode",
      "model": "claude-opus-4-7_high",
      "score": 0.3111111111111111,
      "bestScore": 0.3111111111111111,
      "standardError": 0.09036473223803354,
      "startedAt": "2026-08-12T11:43:31.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mirrorcode-ml-2l-claude-opus-4-7",
      "sourceFile": "mirrorcode.csv"
    },
    {
      "benchmark": "MirrorCode",
      "model": "gemini-3.1-pro-preview_high",
      "score": 0.08888888888888888,
      "bestScore": 0.08888888888888888,
      "standardError": 0.04561939004190448,
      "startedAt": "2026-08-10T16:12:46.343Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mirrorcode-ml-2l-gemini-3-1-pro-preview",
      "sourceFile": "mirrorcode.csv"
    },
    {
      "benchmark": "MirrorCode",
      "model": "gpt-5.5_high",
      "score": 0.1,
      "bestScore": 0.1,
      "standardError": 0.06042179781166438,
      "startedAt": "2026-08-10T16:12:45.865Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mirrorcode-ml-2l-gpt-5-5",
      "sourceFile": "mirrorcode.csv"
    },
    {
      "benchmark": "MirrorCode",
      "model": "gpt-5.4-2026-03-05_high",
      "score": 0.15555555555555556,
      "bestScore": 0.15555555555555556,
      "standardError": 0.0770945044100374,
      "startedAt": "2026-08-10T16:12:45.299Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mirrorcode-ml-2l-gpt-5-4",
      "sourceFile": "mirrorcode.csv"
    },
    {
      "benchmark": "MirrorCode",
      "model": "gpt-5.6-sol_high",
      "score": 0.2,
      "bestScore": 0.2,
      "standardError": 0.09229582069908972,
      "startedAt": "2026-08-10T16:12:44.772Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mirrorcode-ml-2l-gpt-5-6-sol",
      "sourceFile": "mirrorcode.csv"
    },
    {
      "benchmark": "MirrorCode",
      "model": "claude-fable-5_high",
      "score": 0.638888888888889,
      "bestScore": 0.638888888888889,
      "standardError": 0.10404092940368063,
      "startedAt": "2026-08-10T16:12:44.227Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mirrorcode-ml-2l-claude-fable-5",
      "sourceFile": "mirrorcode.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "muse-wonder-1.0_max",
      "score": 0.5416666666666666,
      "bestScore": 0.5416666666666666,
      "standardError": 0.06432138878020231,
      "startedAt": "2026-10-15T22:49:11.000Z",
      "releaseDate": "2026-10-12",
      "organization": "Meta AI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "iuNhWVj65sV54m3cKVwv9x",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "mistral-large-4-0_high",
      "score": 0.20833333333333334,
      "bestScore": 0.20833333333333334,
      "standardError": 0.052199728697406675,
      "startedAt": "2026-10-14T00:39:04.000Z",
      "releaseDate": "2026-10-06",
      "organization": "Mistral AI",
      "country": "France",
      "logViewer": null,
      "sourceId": "cETggiScYbsKGvPW5BAVWJ",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gemini-4-argon_high",
      "score": 0.725,
      "bestScore": 0.725,
      "standardError": 0.056279417919483966,
      "startedAt": "2026-10-13T17:15:28.000Z",
      "releaseDate": "2026-09-30",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bRw53PXBsZXrvWau95hb23",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "deepseek-v4.1-flash_max",
      "score": 0.3416666666666667,
      "bestScore": 0.3416666666666667,
      "standardError": 0.06000431560123094,
      "startedAt": "2026-10-08T19:02:18.000Z",
      "releaseDate": "2026-09-09",
      "organization": "DeepSeek",
      "country": "China",
      "logViewer": null,
      "sourceId": "JFQz7q3hxAKLV5pW6DmkGF",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "claude-haiku-5-5_max",
      "score": 0.475,
      "bestScore": 0.475,
      "standardError": 0.06446761548504554,
      "startedAt": "2026-10-07T20:22:18.000Z",
      "releaseDate": "2026-10-07",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bZMMyetG9pAbKgpxghoTe8",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gpt-6.1-sol_max",
      "score": 0.8,
      "bestScore": 0.8,
      "standardError": 0.04928872620664329,
      "startedAt": "2026-09-29T18:29:32.000Z",
      "releaseDate": "2026-09-29",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "R4vjsSfDrDo6ECQWGTUdWe",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "claude-sonnet-5-5_max",
      "score": 0.75,
      "bestScore": 0.75,
      "standardError": 0.05510647116121488,
      "startedAt": "2026-09-29T15:17:08.000Z",
      "releaseDate": "2026-09-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "5xgoV6NXxdDqscnFNpP2vv",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gpt-6-luna_max",
      "score": 0.44166666666666665,
      "bestScore": 0.44166666666666665,
      "standardError": 0.06410142328406845,
      "startedAt": "2026-09-28T03:08:57.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GVtcaKADdbkuPoydqpKm3E",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "grok-4.5_high",
      "score": 0.225,
      "bestScore": 0.225,
      "standardError": 0.053711144147650755,
      "startedAt": "2026-09-24T17:36:22.000Z",
      "releaseDate": "2026-07-08",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "BFVFWpAaax3RDTXT7apuZE",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "grok-4.6_xhigh",
      "score": 0.4,
      "bestScore": 0.4,
      "standardError": 0.06266221885886569,
      "startedAt": "2026-09-24T17:32:49.000Z",
      "releaseDate": "2026-08-12",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "GM5e6UBbgeBJ83NJe6R6y9",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "grok-4.7_xhigh",
      "score": 0.20833333333333334,
      "bestScore": 0.20833333333333334,
      "standardError": 0.052199728697406675,
      "startedAt": "2026-09-24T16:31:21.000Z",
      "releaseDate": "2026-09-21",
      "organization": "xAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "jWTJKkY9kqYnkvZXCMKnwP",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gpt-6-sol_max",
      "score": 0.5833333333333334,
      "bestScore": 0.5833333333333334,
      "standardError": 0.06418399868645064,
      "startedAt": "2026-09-23T20:34:23.000Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "aV7jRQtNYXjDAcUHhvKxAx",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "claude-opus-5-5_max",
      "score": 0.8333333333333334,
      "bestScore": 0.8333333333333334,
      "standardError": 0.047040472216649226,
      "startedAt": "2026-09-22T16:51:00.000Z",
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4QrkRM7KEDCQvg4wvfhz7N",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "kimi-k3_max",
      "score": 0.3416666666666667,
      "bestScore": 0.3416666666666667,
      "standardError": 0.061169934025019475,
      "startedAt": "2026-09-11T03:00:40.000Z",
      "releaseDate": "2026-07-16",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "fMkb7c7AqFoehCqiy2tLR5",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "kimi-k2.6",
      "score": 0.21666666666666667,
      "bestScore": 0.21666666666666667,
      "standardError": 0.052301098424620544,
      "startedAt": "2026-09-11T02:02:56.000Z",
      "releaseDate": "2026-04-20",
      "organization": "Moonshot",
      "country": "China",
      "logViewer": null,
      "sourceId": "eJcV4jnwhxD6tTxNkrn9Xn",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "qwen3.8-max-0902_xhigh",
      "score": 0.2,
      "bestScore": 0.2,
      "standardError": 0.05207556439232955,
      "startedAt": "2026-09-10T20:24:14.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Alibaba",
      "country": "China",
      "logViewer": null,
      "sourceId": "kpiLkPjDgoKRYp2SiKm9Zo",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gemini-3.7-flash_high",
      "score": 0.26666666666666666,
      "bestScore": 0.26666666666666666,
      "standardError": 0.056331678510110804,
      "startedAt": "2026-09-10T20:23:57.000Z",
      "releaseDate": "2026-08-13",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bqeNkcnEXJeZ9jgmao393V",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gemini-3.6-flash_high",
      "score": 0.23333333333333334,
      "bestScore": 0.23333333333333334,
      "standardError": 0.05376590116477416,
      "startedAt": "2026-09-10T20:23:53.000Z",
      "releaseDate": "2026-07-21",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Kkf7BH3sbSqhEDhhPLcmDN",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gemini-3.8-flash_high",
      "score": 0.31666666666666665,
      "bestScore": 0.31666666666666665,
      "standardError": 0.0593832076297605,
      "startedAt": "2026-09-10T20:23:47.000Z",
      "releaseDate": "2026-09-02",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "Q2QhUc6V3Cu9XS3WZH4tQh",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "claude-fable-5-1_max",
      "score": 0.7,
      "bestScore": 0.7,
      "standardError": 0.05724365605193373,
      "startedAt": "2026-09-10T20:23:45.000Z",
      "releaseDate": "2026-09-01",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9LZMBmKxeBgnf5mvgNvBjW",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gemini-3.1-pro-preview_high",
      "score": 0.26666666666666666,
      "bestScore": 0.26666666666666666,
      "standardError": 0.056331678510110804,
      "startedAt": "2026-09-10T20:23:45.000Z",
      "releaseDate": "2026-02-19",
      "organization": "Google DeepMind",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "NzNcASjLtNrngbNQKcD3G3",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "claude-fable-5_max",
      "score": 0.35833333333333334,
      "bestScore": 0.35833333333333334,
      "standardError": 0.06185876325418331,
      "startedAt": "2026-09-10T20:23:38.000Z",
      "releaseDate": "2026-06-09",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "433Hqw9fVgtRtAKQfareFr",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "claude-opus-5_max",
      "score": 0.6083333333333333,
      "bestScore": 0.6083333333333333,
      "standardError": 0.06185876325418331,
      "startedAt": "2026-09-10T20:23:29.000Z",
      "releaseDate": "2026-07-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "eBTo3dRKwCVuHaRSyN24VA",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "claude-opus-4-8_max",
      "score": 0.425,
      "bestScore": 0.425,
      "standardError": 0.06380695638890652,
      "startedAt": "2026-09-10T20:23:25.000Z",
      "releaseDate": "2026-05-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "mTRwnGKK2QWHrovdPrrMbJ",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "claude-opus-4-7_max",
      "score": 0.3333333333333333,
      "bestScore": 0.3333333333333333,
      "standardError": 0.06137164119322156,
      "startedAt": "2026-09-10T20:23:16.000Z",
      "releaseDate": "2026-04-16",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "NPkiPtHFjjYvdXoG2LLg7A",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "claude-opus-4-6_max",
      "score": 0.2833333333333333,
      "bestScore": 0.2833333333333333,
      "standardError": 0.05866531842313124,
      "startedAt": "2026-09-10T20:22:53.000Z",
      "releaseDate": "2026-02-05",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "4Jq9M5CbuWMyxRHHxavMwA",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "claude-opus-4-5-20251101_64K",
      "score": 0.2833333333333333,
      "bestScore": 0.2833333333333333,
      "standardError": 0.05866531842313124,
      "startedAt": "2026-09-10T20:22:43.000Z",
      "releaseDate": "2025-11-24",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "E3gZb5amR7rcoRkqUGeQRG",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gpt-5.6-luna_max",
      "score": 0.425,
      "bestScore": 0.425,
      "standardError": 0.06380695638890652,
      "startedAt": "2026-09-10T20:22:42.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "UPAoeCG9AhQYyxCGc8C7Eq",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gpt-6-astra_max",
      "score": 0.8,
      "bestScore": 0.8,
      "standardError": 0.04928872620664329,
      "startedAt": "2026-09-10T20:22:39.000Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "BSzNeBCFhFUnZswSqHtAed",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gpt-5.6-sol_max",
      "score": 0.5666666666666667,
      "bestScore": 0.5666666666666667,
      "standardError": 0.06340911315327727,
      "startedAt": "2026-09-10T20:22:35.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "SqvYScyg9JZfiUdNLX6V9M",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gpt-5.6-terra_max",
      "score": 0.5416666666666666,
      "bestScore": 0.5416666666666666,
      "standardError": 0.06432138878020231,
      "startedAt": "2026-09-10T20:22:34.000Z",
      "releaseDate": "2026-07-09",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "dxcm8tfHzGWR8sVJ6kUWMF",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gpt-5.5_xhigh",
      "score": 0.44166666666666665,
      "bestScore": 0.44166666666666665,
      "standardError": 0.06410142328406845,
      "startedAt": "2026-09-10T20:22:27.000Z",
      "releaseDate": "2026-04-23",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "o7rYVQWSXeZPkF2Smwo9h8",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gpt-5.2-2025-12-11_xhigh",
      "score": 0.38333333333333336,
      "bestScore": 0.38333333333333336,
      "standardError": 0.06329764084940143,
      "startedAt": "2026-09-10T20:22:18.000Z",
      "releaseDate": "2025-12-11",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "9te7vbqxpjSaQgHagpaTFg",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "Furniture Assembly",
      "model": "gpt-5.4-2026-03-05_xhigh",
      "score": 0.375,
      "bestScore": 0.375,
      "standardError": 0.06246467928511352,
      "startedAt": "2026-09-10T20:22:11.000Z",
      "releaseDate": "2026-03-05",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "bJ5Pg8JiLzP3Wfoo9M9TxF",
      "sourceFile": "furniture_assembly.csv"
    },
    {
      "benchmark": "SpireClimb",
      "model": "gpt-6.1-sol_max",
      "score": 0.24666666666666667,
      "bestScore": 0.24666666666666667,
      "standardError": 0.03712501046458055,
      "startedAt": "2026-09-30T01:02:59.538Z",
      "releaseDate": "2026-09-29",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "steambench-sts1-a20h-e10776cf8aac6e32+steambench-sts2-a10-9a23c8dea53b6755",
      "sourceFile": "spireclimb.csv"
    },
    {
      "benchmark": "SpireClimb",
      "model": "claude-sonnet-5-5_max",
      "score": 0.013333333333333334,
      "bestScore": 0.013333333333333334,
      "standardError": 0.009364170421755878,
      "startedAt": "2026-09-29T15:20:30.511Z",
      "releaseDate": "2026-09-28",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "steambench-sts1-a20h-adf73a874122aee6+steambench-sts2-a10-5db0e95e99f6b21c",
      "sourceFile": "spireclimb.csv"
    },
    {
      "benchmark": "SpireClimb",
      "model": "gpt-6-luna_max",
      "score": 0,
      "bestScore": 0,
      "standardError": 0,
      "startedAt": "2026-09-28T21:23:58.598Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "steambench-sts1-a20h-fc581bcaa7f17d12+steambench-sts2-a10-0e148062f8e47c4a",
      "sourceFile": "spireclimb.csv"
    },
    {
      "benchmark": "SpireClimb",
      "model": "claude-fable-5-1_max",
      "score": 0.02,
      "bestScore": 0.02,
      "standardError": 0.011389895949029987,
      "startedAt": "2026-09-25T03:54:00.410Z",
      "releaseDate": "2026-09-01",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "steambench-sts1-a20h-3a714bccadb811b3+steambench-sts2-a10-413d1cc930f6d4c6",
      "sourceFile": "spireclimb.csv"
    },
    {
      "benchmark": "SpireClimb",
      "model": "claude-opus-5-5_max",
      "score": 0.03333333333333333,
      "bestScore": 0.03333333333333333,
      "standardError": 0.014498627873361331,
      "startedAt": "2026-09-24T20:02:22.932Z",
      "releaseDate": "2026-09-22",
      "organization": "Anthropic",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "steambench-sts1-a20h-a8e8ab7eb914d4bc+steambench-sts2-a10-ef454c368767b2a0",
      "sourceFile": "spireclimb.csv"
    },
    {
      "benchmark": "SpireClimb",
      "model": "gpt-6-sol_max",
      "score": 0.075,
      "bestScore": 0.075,
      "standardError": 0.021443759811069392,
      "startedAt": "2026-09-24T05:13:21.047Z",
      "releaseDate": "2026-09-22",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "steambench-sts1-a20h-ee8cb2987a2c6c37+steambench-sts2-a10-edacf7bb0ae31edf",
      "sourceFile": "spireclimb.csv"
    },
    {
      "benchmark": "SpireClimb",
      "model": "gpt-6-astra_max",
      "score": 0.25666666666666665,
      "bestScore": 0.25666666666666665,
      "standardError": 0.03679235526872096,
      "startedAt": "2026-09-23T02:12:45.776Z",
      "releaseDate": "2026-09-03",
      "organization": "OpenAI",
      "country": "United States of America",
      "logViewer": null,
      "sourceId": "steambench-sts1-a20h-c8709a73c9fd54e3+steambench-sts2-a10-cdf1c9fa85cb6300",
      "sourceFile": "spireclimb.csv"
    }
  ],
  "coverage": {
    "eciModelGroups": 277,
    "sourceModelVersions": 1149,
    "importedInternalBenchmarks": 15,
    "importedRunRows": 1600,
    "includesAllModels": false,
    "selection": "Every valid model group in the downloaded Epoch Capabilities Index; internal benchmark tables listed in the archive metadata. Externally administered benchmark tables are excluded from this local import."
  },
  "interpretation": "ECI is Epoch’s fitted capability scale, not a percentage or an operational success rate. Scores depend on source coverage and evaluation settings. Confidence bounds are supplied by Epoch. Empty fields remain null. MEKreflect has not rerun these evaluations."
}