{
  "schema": "chi-bench/submission/v1",
  "submission": {
    "id": "kimi-k3-openai-agents",
    "team": "Actava",
    "contact": "dark.savi@gmail.com",
    "agent": "openai-agents",
    "model": "moonshotai/kimi-k3",
    "notes": "Kimi K3 via the OpenAI Agents SDK harness with a 50-turn limit. Cost is derived from the normalized price table rather than provider-reported totals: 71 of 75 trials carry estimated cost and 4 pa_um trials terminated with agent exceptions and report no usage, so leaderboard cost is a lower bound on billed execution cost. Run executed 2026-07-22; filed late (see provenance.filing_note).\n",
    "submitted_at": "2026-08-12T20:20:00Z"
  },
  "dataset": {
    "name": "chi-bench",
    "version": "chi-bench-v1.0.0",
    "domains": [
      "pa_provider",
      "pa_um",
      "cm"
    ]
  },
  "results": {
    "overall": {
      "n_trials": 75,
      "n_tasks": 75,
      "pass_at_1": 0.25333333333333335,
      "mean_cost_usd": 1.27148092,
      "mean_walltime_s": 854.36141916
    },
    "per_domain": {
      "pa_provider": {
        "n_trials": 25,
        "n_tasks": 25,
        "pass_at_1": 0.28,
        "mean_cost_usd": 1.402652808,
        "mean_walltime_s": 592.95700092
      },
      "pa_um": {
        "n_trials": 25,
        "n_tasks": 25,
        "pass_at_1": 0.32,
        "mean_cost_usd": 1.03994136,
        "mean_walltime_s": 792.82956592
      },
      "cm": {
        "n_trials": 25,
        "n_tasks": 25,
        "pass_at_1": 0.16,
        "mean_cost_usd": 1.371848592,
        "mean_walltime_s": 1177.29769064
      }
    },
    "mean_cost_usd": 1.27148092,
    "mean_walltime_s": 854.36141916
  },
  "provenance": {
    "chi_bench_git_sha": "2395e9e2ae7d42cdf2750dc3e67daeb0e672b910",
    "image_digest": null,
    "judge_model": "claude-opus-4-7",
    "harness_version": "0.1.0",
    "code_dirty": false,
    "dataset_version": "chi-bench-v1.0.0",
    "environment": "modal",
    "started_at": "2026-07-22T08:03:23.788878Z",
    "finished_at": "2026-07-22T14:22:38.269923Z",
    "source": "frontier_models_seven_model_full_2026_07 experiment matrix",
    "cost_basis": "All 75 trials have price-table-estimated cost rather than provider-reported cost (cost_reported_trials = 0); 4 pa_um trials with terminal agent exceptions contribute $0.",
    "filing_note": "Run completed 2026-07-22 as part of the 2026-07-24 frontier release wave and was published in the 4/4 results post on 2026-07-29, but the leaderboard submission was not filed at the time. Filed 2026-08-12 with the original, unmodified aggregation.",
    "trajectory_repair": {
      "count": 4,
      "reason": "4 pa_um trials terminated with agent exceptions and completed verification but emitted no trace.jsonl.",
      "method": "Generated ATIF-v1.2 error-only trajectories with the official openai_agents_harness._build_atif_trajectory(records=[], ...) builder from each preserved agent/run_result.json; original logs were not modified."
    }
  }
}