{
  "schema": 1,
  "methodology_version": "3qk7-evidence-index-v0.1",
  "captured_at": "2026-10-20T14:35:00+08:00",
  "notice": "Raw third-party metrics used by the 3qk7 Model Evidence Index. These are not NewLab test results and do not reuse AIHOT aggregate scores or ranks.",
  "sources": [
    {
      "id": "livebench-2",
      "name": "LiveBench 2.0",
      "url": "https://livebench.ai/",
      "role": "objective long-horizon task benchmark",
      "observed_at": "2026-10-20",
      "source_updated": "2026-10-18"
    },
    {
      "id": "artificial-analysis",
      "name": "Artificial Analysis Intelligence Index",
      "url": "https://artificialanalysis.ai/leaderboards/models",
      "role": "independent intelligence benchmark bundle",
      "observed_at": "2026-10-20"
    },
    {
      "id": "agent-arena-work",
      "name": "Agent Arena · Work",
      "url": "https://arena.ai/leaderboard/agent/work/pareto",
      "role": "real agent-session work signal",
      "observed_at": "2026-10-20",
      "source_updated": "2026-10-19"
    }
  ],
  "models": [
    {
      "name": "Claude Fable 5.1",
      "provider": "Anthropic",
      "livebench": {
        "overall": 88,
        "coding": 97.7,
        "reasoning": 91.7,
        "research": 82,
        "data_analysis": 80.3
      },
      "artificial_analysis": 53,
      "agent_arena_work": 14.45
    },
    {
      "name": "Claude Opus 5.5",
      "provider": "Anthropic",
      "livebench": {
        "overall": 86.5,
        "coding": 96.3,
        "reasoning": 92.2,
        "research": 85,
        "data_analysis": 80.3
      },
      "artificial_analysis": 58,
      "agent_arena_work": 19.78
    },
    {
      "name": "GPT-6 Astra",
      "provider": "OpenAI",
      "livebench": {
        "overall": 86.1,
        "coding": 93.7,
        "reasoning": 92.7,
        "research": 85.2,
        "data_analysis": 83
      },
      "artificial_analysis": 53,
      "agent_arena_work": 14.64
    },
    {
      "name": "GPT-6.1 Sol",
      "provider": "OpenAI",
      "livebench": {
        "overall": 85.4,
        "coding": 96,
        "reasoning": 92.6,
        "research": 83.5,
        "data_analysis": 82.7
      },
      "artificial_analysis": 52,
      "agent_arena_work": 13.93
    },
    {
      "name": "Gemini 4 Argon",
      "provider": "Google",
      "livebench": {
        "overall": 84.1,
        "coding": 89.7,
        "reasoning": 95.1,
        "research": 83.1,
        "data_analysis": 85.2
      },
      "artificial_analysis": 53,
      "agent_arena_work": 11.48
    },
    {
      "name": "Claude Fable 5",
      "provider": "Anthropic",
      "livebench": {
        "overall": 83.6,
        "coding": 98.3,
        "reasoning": 89.7,
        "research": 86.9,
        "data_analysis": 80.5
      },
      "artificial_analysis": null,
      "agent_arena_work": 8.45
    },
    {
      "name": "Claude Sonnet 5.5",
      "provider": "Anthropic",
      "livebench": {
        "overall": 82.5,
        "coding": 94.7,
        "reasoning": 91.6,
        "research": 83.6,
        "data_analysis": 59.5
      },
      "artificial_analysis": 56,
      "agent_arena_work": 12.65
    },
    {
      "name": "Claude Opus 5",
      "provider": "Anthropic",
      "livebench": {
        "overall": 82.1,
        "coding": 90.8,
        "reasoning": 91.2,
        "research": 88.9,
        "data_analysis": 74.6
      },
      "artificial_analysis": null,
      "agent_arena_work": 7.7
    },
    {
      "name": "GPT-6 Sol",
      "provider": "OpenAI",
      "livebench": {
        "overall": 81.9,
        "coding": 93.1,
        "reasoning": 88.7,
        "research": 69.3,
        "data_analysis": 81.2
      },
      "artificial_analysis": null,
      "agent_arena_work": 9.47
    },
    {
      "name": "DeepSeek V4.1 Flash 1017",
      "provider": "DeepSeek",
      "livebench": {
        "overall": 79.2,
        "coding": 96,
        "reasoning": 89.1,
        "research": 79.5,
        "data_analysis": 79.5
      },
      "artificial_analysis": 45,
      "agent_arena_work": 6.39
    },
    {
      "name": "Muse Wonder",
      "provider": "Meta",
      "livebench": {
        "overall": 78.1,
        "coding": 86.2,
        "reasoning": 90.6,
        "research": 88,
        "data_analysis": 81.6
      },
      "artificial_analysis": 52,
      "agent_arena_work": 3.73
    }
  ],
  "scoring": {
    "normalization": "Within each raw metric, convert models to midrank percentiles in the current candidate cohort. 0 is cohort bottom, 100 is cohort top.",
    "missing": "Missing metrics receive a neutral 50th-percentile contribution and reduce coverage. Models below 70% weighted coverage are not ranked.",
    "boards": {
      "overall": {
        "LiveBench Overall": 0.5,
        "Artificial Analysis Intelligence": 0.3,
        "Agent Arena Work": 0.2
      },
      "coding": {
        "LiveBench Coding": 0.65,
        "Agent Arena Work": 0.35
      },
      "reasoning": {
        "LiveBench Reasoning": 0.7,
        "Artificial Analysis Intelligence": 0.3
      },
      "professional": {
        "Agent Arena Work": 0.55,
        "LiveBench Research": 0.25,
        "LiveBench Data Analysis": 0.2
      },
      "research": {
        "LiveBench Research": 0.7,
        "Artificial Analysis Intelligence": 0.3
      }
    },
    "exclusions": [
      "Price, API availability, context length and vendor marketing claims do not affect rank."
    ]
  }
}