{
  "schema": "builderwars.eval-source-extract.v1",
  "id": "hle",
  "sourceUrl": "https://lastexam.ai/",
  "sourceSha256": "1ceab46b10ee69777117db9af5851f3e59efd67ff698984f37c1b196bd7e041c",
  "capturedAt": "2026-10-07T07:01:28Z",
  "extraction": "Retained reported score columns; not complete source page or execution evidence",
  "rows": [
    {
      "name": "Gemini 3 Pro",
      "score": 38.3,
      "detail": "Calibration error 57.2% (lower is better)",
      "date": null,
      "rank": 1,
      "sourceRowId": "1"
    },
    {
      "name": "GPT-5",
      "score": 25.3,
      "detail": "Calibration error 50.0% (lower is better)",
      "date": null,
      "rank": 2,
      "sourceRowId": "2"
    },
    {
      "name": "Grok 4",
      "score": 24.5,
      "detail": "Calibration error 56.4% (lower is better)",
      "date": null,
      "rank": 3,
      "sourceRowId": "3"
    },
    {
      "name": "Gemini 2.5 Pro",
      "score": 21.6,
      "detail": "Calibration error 72.0% (lower is better)",
      "date": null,
      "rank": 4,
      "sourceRowId": "4"
    },
    {
      "name": "GPT-5-mini",
      "score": 19.4,
      "detail": "Calibration error 65.0% (lower is better)",
      "date": null,
      "rank": 5,
      "sourceRowId": "5"
    },
    {
      "name": "Claude 4.5 Sonnet",
      "score": 13.7,
      "detail": "Calibration error 65.0% (lower is better)",
      "date": null,
      "rank": 6,
      "sourceRowId": "6"
    },
    {
      "name": "Gemini 2.5 Flash",
      "score": 12.1,
      "detail": "Calibration error 80.0% (lower is better)",
      "date": null,
      "rank": 7,
      "sourceRowId": "7"
    },
    {
      "name": "DeepSeek-R1*",
      "score": 8.5,
      "detail": "Calibration error 73.0% (lower is better)",
      "date": null,
      "rank": 8,
      "sourceRowId": "8"
    },
    {
      "name": "o1",
      "score": 8.0,
      "detail": "Calibration error 83.0% (lower is better)",
      "date": null,
      "rank": 9,
      "sourceRowId": "9"
    },
    {
      "name": "GPT-4o",
      "score": 2.7,
      "detail": "Calibration error 89.0% (lower is better)",
      "date": null,
      "rank": 10,
      "sourceRowId": "10"
    }
  ]
}