{
  "api_version": "1",
  "updated": "2026-10-18",
  "latest_event": "2026-10-17",
  "rights": "© Frontier Log, all rights reserved.",
  "benchmark": {
    "slug": "arc-agi-3",
    "name": "ARC-AGI-3",
    "family": "arc-agi",
    "unit": "%",
    "higher_is_better": true,
    "status": "active",
    "benchmark_url": "https://arcprize.org/arc-agi/3/",
    "introduced": "2026-03-25",
    "creator": "ARC Prize Foundation",
    "introduced_source": "https://arcprize.org/blog/arc-agi-3-launch",
    "settings": [
      {
        "setting": "default",
        "name": "",
        "url": "https://frontierlog.ai/benchmark/arc-agi-3",
        "results": 3
      }
    ]
  },
  "total": 3,
  "results": [
    {
      "id": "gpt-5-6-sol:arc-agi-3:default:1",
      "model": "gpt-5-6-sol",
      "model_name": "GPT-5.6 Sol",
      "model_date": "2026-07-09",
      "event": "https://frontierlog.ai/event/2026-07-09-gpt-5-6",
      "benchmark": "arc-agi-3",
      "benchmark_name": "ARC-AGI-3",
      "setting": "default",
      "setting_name": "",
      "score": 7.78,
      "unit": "%",
      "printed": "7.78%",
      "reported_by": "openai",
      "label": "GPT‑5.6 Sol",
      "source": "https://openai.com/index/gpt-5-6",
      "source_title": "GPT-5.6: Frontier intelligence that scales with your ambition",
      "source_date": "2026-07-09",
      "where": "abstract-reasoning eval table (HTML, last table), row 'ARC-AGI-3⁷', column 'GPT‑5.6 Sol'",
      "own": true,
      "primary": true,
      "parallel": false,
      "notes": "Row carries footnote 7 (superscript dropped); the footnote's text is not in the extracted page (text truncated). The later GPT-6 Astra post prints 7.8% for GPT-5.6 Sol (a later report).",
      "caveat": "",
      "human_verified": true,
      "human_verified_on": "2026-10-18"
    },
    {
      "id": "claude-opus-5:arc-agi-3:default:1",
      "model": "claude-opus-5",
      "model_name": "Claude Opus 5",
      "model_date": "2026-07-24",
      "event": "https://frontierlog.ai/event/2026-07-24-claude-opus-5",
      "benchmark": "arc-agi-3",
      "benchmark_name": "ARC-AGI-3",
      "setting": "default",
      "setting_name": "",
      "score": 30.2,
      "unit": "%",
      "printed": "30.2%",
      "reported_by": "anthropic",
      "label": "Opus 5",
      "source": "https://www.anthropic.com/news/claude-opus-5",
      "source_title": "Introducing Claude Opus 5",
      "source_date": "2026-07-24",
      "where": "benchmark table image, row 'Novel problem-solving / ARC-AGI-3', column 'Opus 5'",
      "own": true,
      "primary": true,
      "parallel": false,
      "notes": "Effort not stated for the table; the 'Novel problem-solving by cost' chart (b5e071ba...png) labels the Opus 5 point 'Opus 5 (high)' with no value. Text: 'Opus 5's score is three times as high as the next-best model'.",
      "caveat": "",
      "human_verified": true,
      "human_verified_on": "2026-10-18",
      "source_image": "https://frontierlog.ai/sources/924165396ca89130.png",
      "source_image_original": "https://www-cdn.anthropic.com/images/4zrzovbb/website/a8fb4f77a9fe240e6f27f3bdc47a137f3c74a29d-2600x2578.png"
    },
    {
      "id": "gpt-6-astra:arc-agi-3:default:1",
      "model": "gpt-6-astra",
      "model_name": "GPT-6 Astra",
      "model_date": "2026-09-03",
      "event": "https://frontierlog.ai/event/2026-09-03-gpt-6-astra",
      "benchmark": "arc-agi-3",
      "benchmark_name": "ARC-AGI-3",
      "setting": "default",
      "setting_name": "",
      "score": 99.9,
      "unit": "%",
      "printed": "99.9%",
      "reported_by": "openai",
      "label": "GPT‑6 Astra",
      "source": "https://openai.com/index/gpt-6-astra",
      "source_title": "GPT-6 Astra: A new generation of intelligence",
      "source_date": "2026-09-03",
      "where": "'Abstract reasoning' eval table (HTML), row 'ARC-AGI-3', column 'GPT‑6 Astra'",
      "own": true,
      "primary": true,
      "parallel": false,
      "notes": "Cell prints '99.9% 1': footnote 1 marker dropped; footnote text not in the extracted page (text truncated). Same value in the intro: 'Astra also saturates ARC-AGI-3 with a 99.9% score'. The post also says Astra 'surpassed our human action-efficiency baseline on 96% of levels' (a different measure, not recorded).",
      "caveat": "",
      "human_verified": true,
      "human_verified_on": "2026-10-18"
    }
  ]
}