{
  "model_name": "Claude Opus 4.6",
  "model_organization": "Anthropic",
  "submitting_organization": "Sierra",
  "submission_date": "2026-05-05",
  "contact_info": {
    "email": "ben.s@sierra.ai",
    "name": "Sierra Research Team"
  },
  "is_new": true,
  "submission_type": "standard",
  "trajectories_available": true,
  "trajectory_files": {
    "banking_knowledge": "claude-opus-4-6_max_banking_knowledge_gpt-5.2_4trials.json"
  },
  "references": [],
  "results": {
    "airline": null,
    "retail": null,
    "telecom": null,
    "banking_knowledge": {
      "pass_1": 27.32,
      "pass_2": 18.73,
      "pass_3": 14.18,
      "pass_4": 11.34,
      "cost": null,
      "retrieval_config": "alltools"
    }
  },
  "reasoning_effort": "max",
  "methodology": {
    "evaluation_date": "2026-05-06",
    "tau2_bench_version": "1.0.1",
    "user_simulator": "gpt-5.2",
    "notes": "Evaluated with reasoning_effort 'max' (the highest effort level this model supports). Retrieval: AllTools (BM25 + dense OpenAI text-embedding-3-large + sandboxed shell). User simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Seed: 300. Banking_knowledge domain only — other domains intentionally excluded from this comparison; the AllTools setting standardizes retrieval across models. banking_knowledge re-graded under tau2-bench v1.0.1 grading fixes on 2026-07-15 (extra-read logging #329, numeric-argument normalization #397, T077-T086 gold trajectories #402); scores prior to v1.0.1 are not comparable.",
    "verification": {
      "modified_prompts": false,
      "omitted_questions": false,
      "details": "Banking_knowledge domain only, AllTools retrieval. Trajectories suppressed for this preliminary AllTools comparison; raw per-task results retained internally."
    }
  },
  "model_release": {
    "release_date": "2026-02-05",
    "announcement_url": null
  }
}