{
  "generated": "2026-09-08",
  "benchmark": "mle-bench",
  "results": [
    {
      "system": "AIDE + o1-preview",
      "developer": "OpenAI",
      "value": 17.12,
      "date": "2024-10-08",
      "source": {
        "url": "https://github.com/openai/mle-bench",
        "kind": "official-leaderboard",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "all",
        "scaffold": "AIDE",
        "notes": "Paper-era baseline (16.9% in abstract, 17.12 +/- 0.61 on the leaderboard); 24h budget."
      }
    },
    {
      "system": "AIRA-dojo + o3",
      "developer": "Meta FAIR",
      "value": 31.6,
      "date": "2025-05-15",
      "source": {
        "url": "https://github.com/openai/mle-bench",
        "kind": "official-leaderboard",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "all",
        "scaffold": "AIRA-dojo",
        "notes": "24h budget."
      }
    },
    {
      "system": "Famou-Agent + Gemini-2.5-Pro",
      "developer": "Baidu",
      "value": 43.56,
      "date": "2025-10-10",
      "source": {
        "url": "https://github.com/openai/mle-bench",
        "kind": "official-leaderboard",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "all",
        "scaffold": "Famou-Agent",
        "notes": "24h budget."
      }
    },
    {
      "system": "Famou-Agent 2.0 + Gemini-3-Pro-Preview",
      "developer": "Baidu",
      "value": 64.44,
      "date": "2026-02-23",
      "source": {
        "url": "https://github.com/openai/mle-bench",
        "kind": "official-leaderboard",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "all",
        "scaffold": "Famou-Agent 2.0",
        "notes": "Top of the main leaderboard as of access date; 24h budget."
      }
    },
    {
      "system": "AIBuildAI + Claude-Opus-4.6",
      "developer": "AIBuildAI",
      "value": 63.11,
      "date": "2026-03-06",
      "source": {
        "url": "https://github.com/openai/mle-bench",
        "kind": "official-leaderboard",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "all",
        "scaffold": "AIBuildAI",
        "notes": "24h budget."
      }
    }
  ]
}