eval-run:gpqa-diamond.gemini-3-pro.2025-11-18

{
  "id": "eval-run:gpqa-diamond.gemini-3-pro.2025-11-18",
  "_kind": "EvalRun",
  "_file": "benchmarks/eval-runs/eval-runs-google.yaml",
  "_cluster": "benchmarks",
  "attributes": {
    "target": "model:gemini-3-pro@current",
    "benchmarkId": "benchmark:gpqa",
    "testSetId": "test-set:gpqa-diamond-2024",
    "targetId": "model:gemini-3-pro@current",
    "runAt": "2025-11-18T00:00:00Z",
    "runBy": "google-deepmind",
    "configHash": "sha256:google-gemini-3-pro-gpqa-diamond-2025-11-18"
  },
  "outgoingEdges": [
    {
      "from": "eval-run:gpqa-diamond.gemini-3-pro.2025-11-18",
      "to": "model:gemini-3-pro@current",
      "kind": "evaluates_target"
    },
    {
      "from": "eval-run:gpqa-diamond.gemini-3-pro.2025-11-18",
      "to": "test-set:gpqa-diamond-2024",
      "kind": "uses_test_set"
    },
    {
      "from": "eval-run:gpqa-diamond.gemini-3-pro.2025-11-18",
      "to": "benchmark:gpqa",
      "kind": "for_benchmark"
    }
  ],
  "incomingEdges": [
    {
      "from": "eval-result:gpqa-diamond.gemini-3-pro.2025-11-18.accuracy",
      "to": "eval-run:gpqa-diamond.gemini-3-pro.2025-11-18",
      "kind": "belongs_to_eval_run"
    }
  ]
}