{
  "schema_version": "sensebench-run-v2",
  "run_id": "gemini-2.5-flash-p001-lexen-v1-20260614",
  "created_at": "2026-06-14T11:03:39.454582+00:00",
  "git_commit": "2aac21b87b716ba75c3662d6630af4eca430f1b7",
  "runner": {
    "github_handle": "vassiliphilippov",
    "name": "Vassili Philippov",
    "contact": null
  },
  "dataset": {
    "dataset_id": "lexen",
    "dataset_version": "lexen-v1",
    "dataset_revision": null,
    "content_hash": "sha256:5fd4382b93f19087a1e31f6dd7d1db17c1eb17ff80fcbe1d3fdd55c0c3ecefe8",
    "item_count": 4861
  },
  "prompt": {
    "id": "p001",
    "sensebench_version": "0.1.0"
  },
  "model": {
    "kind": "cloud_llm",
    "display_name": "gemini/gemini-2.5-flash",
    "requested_model": "gemini/gemini-2.5-flash",
    "resolved_model": "gemini-2.5-flash",
    "resolved_model_counts": {
      "gemini-2.5-flash": 4971
    },
    "llm_vendor": "Google",
    "api_provider": "Google",
    "source_kind": "proprietary",
    "license": null,
    "model_url": null,
    "reasoning_effort": null,
    "endpoint_base_url": null
  },
  "sampling": {
    "temperature": 0.0,
    "top_p": null,
    "max_tokens": 2048,
    "seed": null,
    "extra": {}
  },
  "policy": {
    "votes_per_item": 1,
    "semantic_reasks_per_invalid_vote": 1,
    "tie_break": "earliest_vote",
    "monosemous_policy": "short_circuit"
  },
  "machine": null,
  "execution": {
    "concurrency": 32,
    "warmup_call_count": 0,
    "timing": {
      "benchmark_started_at": "2026-06-14T10:51:23.219330+00:00",
      "benchmark_ended_at": "2026-06-14T11:03:39.452310+00:00",
      "benchmark_seconds": 736.2276470420184,
      "setup_seconds": 1.043500917032361
    }
  },
  "totals": {
    "item_count": 4861,
    "correct_count": 4346,
    "accuracy": 0.8940547212507715,
    "call_count": 4975,
    "usage": {
      "input_tokens": 3215331,
      "cached_input_tokens": 12087,
      "output_tokens": 4533437,
      "reasoning_output_tokens": 4474968
    },
    "cost": {
      "currency": "USD",
      "total_usd": 12.294928310000001,
      "input_uncached_usd": 0.9609732,
      "input_cached_usd": 0.00036260999999999997,
      "output_usd": 11.333592500000002,
      "input_uncached_unit_price_usd": 3e-7,
      "input_cached_unit_price_usd": 3e-8,
      "output_unit_price_usd": 2.5e-6,
      "source": "litellm_estimate"
    },
    "elapsed_seconds": 736.2276470420184
  }
}