[
  {
    "benchmark": "swe-bench",
    "score": 74.6,
    "metric": "accuracy",
    "cost_per_instance": 0.86,
    "average_runtime": 476.0,
    "full_archive": "https://results.eval.all-hands.dev/eval-21010530639-gpt-5-2_litellm_proxy-openai-gpt-5-2-2025-12-11_26-01-15-04-24.tar.gz",
    "tags": [
      "swe-bench"
    ],
    "agent_version": "v1.8.3",
    "submission_time": "2026-01-26T15:55:26.395894+00:00"
  },
  {
    "benchmark": "gaia",
    "score": 65.5,
    "metric": "accuracy",
    "cost_per_instance": 0.48,
    "average_runtime": 189.0,
    "full_archive": "https://results.eval.all-hands.dev/eval-21041979432-gpt-5-2_litellm_proxy-openai-gpt-5-2-2025-12-11_26-01-15-20-40.tar.gz",
    "tags": [
      "gaia"
    ],
    "agent_version": "v1.8.3",
    "submission_time": "2026-01-26T15:55:26.395894+00:00"
  },
  {
    "benchmark": "commit0",
    "score": 50.0,
    "metric": "accuracy",
    "cost_per_instance": 4.7,
    "average_runtime": 1580.0,
    "full_archive": "https://results.eval.all-hands.dev/commit0/litellm_proxy-openai-gpt-5-2-2025-12-11/25033142930/results.tar.gz",
    "tags": [
      "commit0"
    ],
    "agent_version": "v1.17.0",
    "submission_time": "2026-04-28T08:17:00+00:00",
    "eval_visualization_page": "https://laminar.sh/shared/evals/c343765e-1fdd-494f-96b7-de843c272a36"
  },
  {
    "benchmark": "swt-bench",
    "score": 73.2,
    "metric": "accuracy",
    "cost_per_instance": 0.56,
    "average_runtime": 347.0,
    "full_archive": "https://results.eval.all-hands.dev/eval-21146218648-gpt-5-2_litellm_proxy-openai-gpt-5-2-2025-12-11_26-01-19-23-58.tar.gz",
    "tags": [
      "swt-bench"
    ],
    "agent_version": "v1.8.3",
    "submission_time": "2026-01-26T15:55:26.395894+00:00"
  },
  {
    "benchmark": "swe-bench-multimodal",
    "score": 30.9,
    "metric": "solveable_accuracy",
    "cost_per_instance": 2.77,
    "average_runtime": 1571.0,
    "full_archive": "https://results.eval.all-hands.dev/eval-21320837315-gpt-5-2_litellm_proxy-openai-gpt-5-2-2025-12-11_26-01-25-04-55.tar.gz",
    "tags": [
      "swe-bench-multimodal"
    ],
    "component_scores": {
      "solveable_accuracy": 30.9,
      "unsolveable_accuracy": 0.0,
      "combined_accuracy": 20.6,
      "solveable_resolved": 21,
      "solveable_total": 68,
      "unsolveable_resolved": 0,
      "unsolveable_total": 34
    },
    "agent_version": "v1.8.3",
    "submission_time": "2026-01-26T15:55:26.395894+00:00"
  }
]