{
  "experiment_id": "weekly-board",
  "calculated_at": "2026-08-22T15:48:07.400963",
  "benchmark_version": "0.3.0",
  "grading_basis": "Saved GradeResult success decisions; API errors count as failures.",
  "grader_implementation_version": null,
  "grading_version_gap": "Snapshots do not record a separate grader implementation version. Benchmark version is not proof of a grader code revision.",
  "answer_present_diagnostic": "content_successes also counts expected answer tokens found in rejected responses; it does not regrade correctness or define the published cost ranking.",
  "authored_tasks": 32,
  "attempted_configurations": 34,
  "scored_configurations": 33,
  "total_attempts": 3264,
  "included_attempts": 3168,
  "excluded_attempts": 96,
  "excluded_configurations": [
    {
      "model_key": "or:muse-spark-1.2",
      "attempts": 96,
      "reason": "No successful or billable response recorded; capability unavailable."
    }
  ],
  "cost_basis": "billed",
  "pricing_versions_by_configuration": {
    "anthropic:haiku-4.5": "2026-07-09",
    "anthropic:opus-4.6": "2026-07-09",
    "anthropic:sonnet-4.6": "2026-07-09",
    "or:deepseek-v3.2": "2026-08-22",
    "or:deepseek-v4-flash": "2026-08-12",
    "or:deepseek-v4-pro": "2026-08-12",
    "or:gemini-2.5-flash": "2026-07-26",
    "or:gemini-3.6-flash": "2026-08-12",
    "or:gemini-3.7-flash": "2026-08-21",
    "or:glm-4.7-flash": "2026-07-26",
    "or:glm-5": "2026-08-15",
    "or:glm-5.2": "2026-08-19",
    "or:glm-5.3": "2026-08-21",
    "or:gpt-5.4-mini": "2026-07-26",
    "or:gpt-5.6-luna": "2026-08-18",
    "or:gpt-5.6-luna-pro": "2026-08-18",
    "or:gpt-5.6-sol": "2026-08-22",
    "or:gpt-5.6-sol-pro": "2026-08-12",
    "or:gpt-5.6-terra": "2026-08-18",
    "or:gpt-5.6-terra-pro": "2026-08-12",
    "or:gpt-oss-120b": "2026-08-12",
    "or:grok-4.5": "2026-08-12",
    "or:grok-4.6": "2026-08-12",
    "or:kimi-k2.5": "2026-08-19",
    "or:kimi-k3": "2026-08-01",
    "or:llama-4-maverick": "2026-08-15",
    "or:muse-spark-1.2": "2026-08-21",
    "or:opus-5-fast": "2026-08-12",
    "or:qwen3.5-flash": "2026-07-26",
    "or:qwen3.8-2.4t": "2026-08-12",
    "or:qwen3.8-27b": "2026-08-21",
    "or:qwen3.8-max": "2026-08-01",
    "or:seed-2.0-code": "2026-08-21",
    "or:seed-2.1-turbo": "2026-08-21"
  },
  "aggregation": {
    "task": "total attempt cost / saved accepted outputs",
    "vertical": "sum(task weight * task cost) / sum(task weight * task successes)",
    "overall": "weighted mean of available vertical cost-per-success rates; zero-success verticals are unavailable and omitted, not pooled cost/successes",
    "vertical_weights": {
      "legal": 1.0,
      "medical": 1.0,
      "customer_service": 1.0,
      "banking": 1.0,
      "coding": 1.0
    }
  },
  "headline_spread": {
    "basis": "Highest / lowest available overall weighted cost per saved accepted output",
    "lowest": {
      "model_key": "or:gpt-oss-120b",
      "cost_per_success_usd": 7.0398157997558e-05,
      "successes": 92,
      "attempts": 96
    },
    "highest": {
      "model_key": "or:gpt-5.6-sol-pro",
      "cost_per_success_usd": 0.019206325713754843,
      "successes": 93,
      "attempts": 96
    },
    "ratio": 272.8242650101879
  },
  "scope": "Pilot of authored short single-prompt tasks; no full production agent evaluation. API costs exclude human review, tools, integration and downstream error costs."
}