{
  "benchmark": "postcutoff_pilot_v1",
  "date": "2026-10-05",
  "task_count": 5,
  "dataset_sha256": "b5ef06357d5c0077ed5dbf35508eeee46a73dfb22dbcc81f7510989601ca4595",
  "standalone": [
    {
      "name": "Mistral Small 4",
      "correct": 2,
      "total": 5,
      "accuracy_pct": 40.0,
      "api_cost_usd": 0.077808951,
      "mean_wall_s": 119.73220639179927
    },
    {
      "name": "Mistral Large 3",
      "correct": 2,
      "total": 5,
      "accuracy_pct": 40.0,
      "api_cost_usd": 0.06653625,
      "mean_wall_s": 75.65426251639728
    },
    {
      "name": "Mistral Medium 3.5",
      "correct": 2,
      "total": 5,
      "accuracy_pct": 40.0,
      "api_cost_usd": 0.99632115,
      "mean_wall_s": 120.7311589667981
    }
  ],
  "fusion": {
    "name": "ALEXIS Fusion (Mistral-only)",
    "correct": 4,
    "total": 5,
    "accuracy_pct": 80.0,
    "api_cost_usd": 0.10555177600000001,
    "mean_wall_s": 36.26889617500128
  },
  "control": {
    "name": "Mistral Large 3 with computation",
    "correct": 3,
    "total": 5,
    "accuracy_pct": 60.0,
    "api_cost_usd": 0.01788259,
    "mean_wall_s": 19.912859583389945
  },
  "method": "Same five published questions and final numeric scoring; one run per question per configuration. Fusion and the additional control use computation tools. Tools and settings differ from standalone models.",
  "cost_basis": "Total metered API cost in USD for five questions, including all model calls. Setup calls, local compute, hosting, taxes and fees excluded.",
  "time_basis": "Mean elapsed wall time per question, including orchestration, network and program execution.",
  "scope": "Small repeat pilot on public tasks, not an unseen evaluation or an isolated test of fusion alone."
}