{
  "generated": "2026-09-08",
  "benchmark": "gdpval",
  "results": [
    {
      "system": "GPT-4o",
      "developer": "OpenAI",
      "value": 12.5,
      "date": "2025-10-05",
      "source": {
        "url": "https://arxiv.org/abs/2510.04374",
        "kind": "paper",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "gold",
        "notes": "Table 2 of the paper; expert pairwise grading, 220 gold tasks."
      }
    },
    {
      "system": "o3",
      "developer": "OpenAI",
      "value": 35.2,
      "date": "2025-10-05",
      "source": {
        "url": "https://arxiv.org/abs/2510.04374",
        "kind": "paper",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "gold",
        "notes": "Table 2 of the paper; expert pairwise grading."
      }
    },
    {
      "system": "GPT-5 (high)",
      "developer": "OpenAI",
      "value": 39.0,
      "date": "2025-10-05",
      "source": {
        "url": "https://arxiv.org/abs/2510.04374",
        "kind": "paper",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "gold",
        "reasoning_effort": "high",
        "tools": true,
        "notes": "Table 2; web search and code interpreter enabled. OpenAI's 2025-12 post cites 38.8% for the same model."
      }
    },
    {
      "system": "Claude Opus 4.1",
      "developer": "Anthropic",
      "value": 47.6,
      "date": "2025-10-05",
      "source": {
        "url": "https://arxiv.org/abs/2510.04374",
        "kind": "paper",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "gold",
        "notes": "Best model in the paper; sampled via the Claude UI with file-creation features. Run by OpenAI, not Anthropic."
      }
    },
    {
      "system": "GPT-5.2 Thinking",
      "developer": "OpenAI",
      "value": 70.9,
      "date": "2025-12-11",
      "source": {
        "url": "https://openai.com/index/introducing-gpt-5-2/",
        "kind": "developer-report",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "gold",
        "reasoning_effort": "heavy",
        "tools": true,
        "notes": "Wins-or-ties with expert human judges; run with ChatGPT spreadsheet/presentation tools not available to GPT-5."
      }
    },
    {
      "system": "GPT-5.2 Pro",
      "developer": "OpenAI",
      "value": 74.1,
      "date": "2025-12-11",
      "source": {
        "url": "https://openai.com/index/introducing-gpt-5-2/",
        "kind": "developer-report",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "gold",
        "tools": true,
        "notes": "Wins-or-ties, from the appendix table of the GPT-5.2 announcement."
      }
    }
  ]
}