{
  "schema": "cortex.benchmark.claims.v1",
  "slug": "tau2-open-model-router",
  "benchmark": "tau2-bench",
  "split": "base",
  "as_of": "2026-08-02",
  "estimand": "paired common-task point comparison",
  "comparison": {
    "router": {
      "tasks": 99,
      "passed": 80,
      "quality": 0.8080808080808081,
      "cost": 0.15942909696969695
    },
    "baseline": {
      "tasks": 99,
      "passed": 86,
      "quality": 0.8686868686868687,
      "cost": 0.41284039393939403
    },
    "quality_delta_percentage_points": -6.060606060606066,
    "cost_reduction_percent": 61.382389099962566
  },
  "full_arm_metrics": [
    {
      "id": "qwen",
      "label": "Qwen3.7 Plus",
      "tasks": 100,
      "passed": 20,
      "quality": 0.2,
      "cost": 0.10310672800000002
    },
    {
      "id": "glm",
      "label": "GLM-5.2",
      "tasks": 100,
      "passed": 83,
      "quality": 0.83,
      "cost": 0.307022752
    },
    {
      "id": "kimi",
      "label": "Kimi K3",
      "tasks": 99,
      "passed": 86,
      "quality": 0.8686868686868687,
      "cost": 0.41284039393939403
    },
    {
      "id": "router",
      "label": "Cortex task-level router",
      "tasks": 100,
      "passed": 81,
      "quality": 0.81,
      "cost": 0.15928938199999998
    }
  ],
  "route_counts": {
    "qwen": 5,
    "glm": 44,
    "kimi": 51
  },
  "published_rows_sha256": "5597703dd8171f741ad92153009d2b6197ccb23c4d959cb2d5873304403bf947",
  "provenance": {
    "source_artifact": "tau2_codex_2arm_100",
    "source_artifact_sha256": "not recorded in supplied artifact",
    "router_build_id": "not recorded in supplied artifact",
    "route_score_provenance": "selected_arm imported from sticky_router_per_task.csv; published arm labels follow sticky summary_report series rename (router occupies former mid-tier series; mid-tier carries former routed aggregate). Decision mix still uses sticky selected_arm.",
    "benchmark_version": "not recorded in supplied artifact",
    "reward_implementation_version": "not recorded in supplied artifact",
    "pricing_source": "Costs use agent_cost recorded by the benchmark harness; the underlying price table was not included in the supplied artifact.",
    "pricing_as_of": "2026-08-02",
    "held_out_status": "The supplied artifact does not establish whether this cohort was held out from router or gate tuning."
  }
}