{
  "checkedAt": "2026-10-01",
  "schemaVersion": 1,
  "efforts": [
    "low",
    "medium",
    "high",
    "xhigh",
    "max"
  ],
  "models": {
    "sol": {
      "name": "GPT-6.1 Sol",
      "released": "2026-09-29",
      "ultra": "Eligible Codex / Work surfaces; no published comparable benchmark point",
      "prices": {
        "input": 2,
        "cached": 0.1,
        "write": 2.5,
        "output": 10
      }
    },
    "luna": {
      "name": "GPT-6 Luna",
      "released": "2026-09-22",
      "ultra": "Not supported",
      "prices": {
        "input": 0.1,
        "cached": 0.01,
        "write": 0.125,
        "output": 0.5
      }
    }
  },
  "sources": {
    "sol": "https://openai.com/index/introducing-gpt-6-1-sol/",
    "luna": "https://openai.com/index/introducing-gpt-6-sol-and-luna/",
    "pricing": "https://developers.openai.com/api/docs/pricing",
    "aa": "https://artificialanalysis.ai/leaderboards/models/",
    "aaMethod": "https://artificialanalysis.ai/methodology/"
  },
  "benchmarks": {
    "coding": {
      "name": "DeepSWE v1.1",
      "kind": "official",
      "published": "2026-09-29 / 2026-09-22",
      "unit": "%",
      "note": "OpenAI release snapshot; cost per attempted benchmark task. Not a live service measurement.",
      "points": [
        {
          "model": "sol",
          "effort": "low",
          "score": 64.38,
          "cost": 0.1714
        },
        {
          "model": "sol",
          "effort": "medium",
          "score": 73.01,
          "cost": 0.4196
        },
        {
          "model": "sol",
          "effort": "high",
          "score": 75.22,
          "cost": 0.6461
        },
        {
          "model": "sol",
          "effort": "xhigh",
          "score": 71.9,
          "cost": 0.7886
        },
        {
          "model": "sol",
          "effort": "max",
          "score": 71.9,
          "cost": 1.5711
        },
        {
          "model": "luna",
          "effort": "low",
          "score": 2.43,
          "cost": 0.0057
        },
        {
          "model": "luna",
          "effort": "medium",
          "score": 44.47,
          "cost": 0.0518
        },
        {
          "model": "luna",
          "effort": "high",
          "score": 59.29,
          "cost": 0.0838
        },
        {
          "model": "luna",
          "effort": "xhigh",
          "score": 61.28,
          "cost": 0.1096
        },
        {
          "model": "luna",
          "effort": "max",
          "score": 66.59,
          "cost": 0.2169
        }
      ]
    },
    "automation": {
      "name": "AutomationBench 1.0.6",
      "kind": "official",
      "published": "2026-09-29 / 2026-09-22",
      "unit": "%",
      "note": "OpenAI release snapshot; cost per attempted benchmark task. Not a live service measurement.",
      "points": [
        {
          "model": "sol",
          "effort": "low",
          "score": 24.7,
          "cost": 0.157
        },
        {
          "model": "sol",
          "effort": "medium",
          "score": 31.7,
          "cost": 0.1917
        },
        {
          "model": "sol",
          "effort": "high",
          "score": 33.2,
          "cost": 0.2255
        },
        {
          "model": "sol",
          "effort": "xhigh",
          "score": 35.5,
          "cost": 0.2508
        },
        {
          "model": "sol",
          "effort": "max",
          "score": 36.1,
          "cost": 0.2989
        },
        {
          "model": "luna",
          "effort": "low",
          "score": 1.2,
          "cost": 0.006
        },
        {
          "model": "luna",
          "effort": "medium",
          "score": 9.4,
          "cost": 0.0163
        },
        {
          "model": "luna",
          "effort": "high",
          "score": 14.5,
          "cost": 0.0208
        },
        {
          "model": "luna",
          "effort": "xhigh",
          "score": 12.6,
          "cost": 0.0245
        },
        {
          "model": "luna",
          "effort": "max",
          "score": 20.7,
          "cost": 0.0367
        }
      ]
    },
    "computer": {
      "name": "OSWorld 2.0 offline",
      "kind": "official",
      "published": "2026-09-29 / 2026-09-22",
      "unit": "%",
      "note": "OpenAI release snapshot; cost per attempted benchmark task. Not a live service measurement. Luna values predate the September 25 image encoding fix; this is a historical comparison.",
      "points": [
        {
          "model": "sol",
          "effort": "low",
          "score": 58.96,
          "cost": 0.4248
        },
        {
          "model": "sol",
          "effort": "medium",
          "score": 66.84,
          "cost": 0.7675
        },
        {
          "model": "sol",
          "effort": "high",
          "score": 69.56,
          "cost": 0.9613
        },
        {
          "model": "sol",
          "effort": "xhigh",
          "score": 69.38,
          "cost": 1.0466
        },
        {
          "model": "sol",
          "effort": "max",
          "score": 71.42,
          "cost": 1.2689
        },
        {
          "model": "luna",
          "effort": "low",
          "score": 8.26,
          "cost": 0.0297
        },
        {
          "model": "luna",
          "effort": "medium",
          "score": 31.54,
          "cost": 0.0624
        },
        {
          "model": "luna",
          "effort": "high",
          "score": 41.43,
          "cost": 0.1227
        },
        {
          "model": "luna",
          "effort": "xhigh",
          "score": 46.7,
          "cost": 0.1639
        },
        {
          "model": "luna",
          "effort": "max",
          "score": 52.68,
          "cost": 0.2678
        }
      ]
    },
    "average": {
      "name": "Shared-task average (custom)",
      "kind": "derived",
      "published": "Launch snapshots",
      "unit": "%",
      "note": "Equal mean of DeepSWE, AutomationBench and OSWorld scores and task costs. Mixed task families; not an official software engineering index.",
      "points": [
        {
          "model": "sol",
          "effort": "low",
          "score": 49.34666666666667,
          "cost": 0.25106666666666666
        },
        {
          "model": "sol",
          "effort": "medium",
          "score": 57.18333333333333,
          "cost": 0.4596
        },
        {
          "model": "sol",
          "effort": "high",
          "score": 59.326666666666675,
          "cost": 0.6109666666666667
        },
        {
          "model": "sol",
          "effort": "xhigh",
          "score": 58.926666666666655,
          "cost": 0.6953333333333332
        },
        {
          "model": "sol",
          "effort": "max",
          "score": 59.806666666666665,
          "cost": 1.0463
        },
        {
          "model": "luna",
          "effort": "low",
          "score": 3.9633333333333334,
          "cost": 0.0138
        },
        {
          "model": "luna",
          "effort": "medium",
          "score": 28.47,
          "cost": 0.043500000000000004
        },
        {
          "model": "luna",
          "effort": "high",
          "score": 38.406666666666666,
          "cost": 0.07576666666666666
        },
        {
          "model": "luna",
          "effort": "xhigh",
          "score": 40.193333333333335,
          "cost": 0.09933333333333333
        },
        {
          "model": "luna",
          "effort": "max",
          "score": 46.656666666666666,
          "cost": 0.17379999999999998
        }
      ]
    },
    "independent": {
      "name": "Artificial Analysis Intelligence Index",
      "kind": "independent",
      "published": "Checked 2026-10-01; measurement dates not exposed in table",
      "unit": "index",
      "note": "Single leaderboard snapshot; broad intelligence, not a coding average. Published task costs are rounded. Other AA pages returned differing values; do not merge their snapshots. Speed assumes the source\u2019s standardized 500-answer-token workload, not completion of a software task.",
      "points": [
        {
          "model": "sol",
          "effort": "low",
          "score": 42,
          "cost": 0.13,
          "throughput": 60,
          "first": 2.61,
          "total": 10.99
        },
        {
          "model": "sol",
          "effort": "medium",
          "score": 48,
          "cost": 0.21,
          "throughput": 59,
          "first": 5.29,
          "total": 13.71
        },
        {
          "model": "sol",
          "effort": "high",
          "score": 50,
          "cost": 0.32,
          "throughput": 63,
          "first": 57.26,
          "total": 65.13
        },
        {
          "model": "sol",
          "effort": "xhigh",
          "score": 51,
          "cost": 0.39,
          "throughput": 63,
          "first": 107.76,
          "total": 115.66
        },
        {
          "model": "sol",
          "effort": "max",
          "score": 52,
          "cost": 0.72,
          "throughput": 65,
          "first": 281.94,
          "total": 289.6
        },
        {
          "model": "luna",
          "effort": "low",
          "score": 21,
          "cost": 0.0045,
          "throughput": 134,
          "first": 2.45,
          "total": 6.18
        },
        {
          "model": "luna",
          "effort": "medium",
          "score": 29,
          "cost": 0.02,
          "throughput": null,
          "first": null,
          "total": null
        },
        {
          "model": "luna",
          "effort": "high",
          "score": 32,
          "cost": 0.03,
          "throughput": 123,
          "first": 12,
          "total": 16.06
        },
        {
          "model": "luna",
          "effort": "xhigh",
          "score": 34,
          "cost": 0.04,
          "throughput": 131,
          "first": 20.86,
          "total": 24.67
        },
        {
          "model": "luna",
          "effort": "max",
          "score": 37,
          "cost": 0.07,
          "throughput": 127,
          "first": 123.89,
          "total": 127.82
        }
      ]
    }
  },
  "history": [
    {
      "date": "2026-10-01",
      "note": "Initial reviewed snapshot. OpenAI embedded chart precision retained; current API prices checked independently. Artificial Analysis leaderboard isolated from comparison-page discrepancies."
    },
    {
      "date": "2026-09-25",
      "note": "OpenAI changelog: Luna computer-use image encoding fix. September 22 launch OSWorld values predate this change."
    }
  ]
}