{
  "schema": "chi-bench/submission/v1",
  "submission": {
    "id": "inkling-256k-openai-agents",
    "team": "Actava",
    "contact": "dark.savi@gmail.com",
    "agent": "openai-agents",
    "model": "thinkingmachines/Inkling:peft:262144",
    "notes": "Inkling 256K via the OpenAI Agents SDK harness and Tinker Chat Completions, with high reasoning effort and a 50-turn limit. Reported cost reflects recorded token usage; 25 terminal agent failures reported zero usage and therefore contribute $0 to the aggregate cost.\n",
    "submitted_at": "2026-07-24T23:02:07Z"
  },
  "dataset": {
    "version": "chi-bench-v1.0.0",
    "domains": [
      "pa_provider",
      "pa_um",
      "cm"
    ],
    "name": "chi-bench"
  },
  "results": {
    "overall": {
      "n_trials": 75,
      "n_tasks": 75,
      "pass_at_1": 0.08,
      "mean_cost_usd": 0.9234470439466665,
      "mean_walltime_s": 140.97706868
    },
    "per_domain": {
      "pa_provider": {
        "n_trials": 25,
        "n_tasks": 25,
        "pass_at_1": 0.04,
        "mean_cost_usd": 0.8389722792,
        "mean_walltime_s": 121.97677540000001
      },
      "pa_um": {
        "n_trials": 25,
        "n_tasks": 25,
        "pass_at_1": 0.16,
        "mean_cost_usd": 1.4461965891200002,
        "mean_walltime_s": 213.6378354
      },
      "cm": {
        "n_trials": 25,
        "n_tasks": 25,
        "pass_at_1": 0.04,
        "mean_cost_usd": 0.48517226352000004,
        "mean_walltime_s": 87.31659524000001
      }
    },
    "mean_cost_usd": 0.9234470439466665,
    "mean_walltime_s": 140.97706868
  },
  "provenance": {
    "chi_bench_git_sha": "4f39c1e77a90c28cd28e3f2264a8389a0dede039",
    "image_digest": null,
    "judge_model": "claude-opus-4-7",
    "harness_version": "0.1.0",
    "code_dirty": true,
    "dataset_version": "chi-bench-v1.0.0",
    "environment": "modal",
    "started_at": "2026-07-22T14:25:50.976659Z",
    "finished_at": "2026-07-22T16:53:02.466832Z",
    "source": "frontier_models_full_2026_07 experiment matrix",
    "trajectory_repair": {
      "count": 25,
      "reason": "Terminal agent failures 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."
    },
    "recorded_cost_caveat": "The 25 terminal failures with error-only trajectories reported zero input/output/cache tokens and null cost; leaderboard cost is therefore a lower bound on billed execution cost."
  }
}