{
  "published_at": "2026-07-19",
  "benchmark": "jcode bench v1",
  "benchmark_commit": "a9bfcdd9ed6cba355bef1025b552ee3da70ce2c0",
  "score_semantics": "score = log2(given_cost / optimized_cost); higher is better",
  "harness": {
    "name": "OpenCode",
    "version": "1.0.203",
    "execution": "solo",
    "subagents_enabled": false,
    "explicit_subagent_events": 0
  },
  "environment": {
    "prompt": "historical benchmark prompt",
    "reasoning_effort": "high",
    "cpus": 4,
    "memory_mib": 8192,
    "single_use_containers": true,
    "runtime": "Modal"
  },
  "models": {
    "claude-opus-4-8": {
      "provider": "anthropic",
      "model": "claude-opus-4-8",
      "reasoning_effort": "high",
      "tasks": {
        "json-unescape": {
          "run_id": "20260718T203139Z-opencode-opus48-json-unescape",
          "final_score": 1.9991,
          "best_score": 1.9991,
          "grade_count": 83,
          "active_seconds": 2562.78,
          "correctness_gate": "PASS",
          "subagent_events": 0
        },
        "float-print": {
          "run_id": "20260718T211537Z-opencode-opus48-float-print",
          "final_score": 7.204,
          "best_score": 7.2077,
          "grade_count": 33,
          "active_seconds": 5715.091,
          "correctness_gate": "PASS",
          "subagent_events": 0
        },
        "utf16-transcode": {
          "run_id": "20260718T211537Z-opencode-opus48-utf16-transcode",
          "final_score": 1.8526,
          "best_score": 1.8638,
          "grade_count": 42,
          "active_seconds": 1890.82,
          "correctness_gate": "PASS",
          "subagent_events": 0
        }
      },
      "aggregate": {
        "mean_final_score": 3.6852,
        "mean_best_score": 3.6902,
        "successful_active_seconds": 10168.691,
        "successful_active_minutes": 169.478,
        "completed_tasks": 3,
        "subagent_events": 0
      }
    }
  },
  "historical_jcode_context": {
    "jcode_opus_mean_best_score": 4.869033,
    "opencode_opus_mean_best_score": 3.6902,
    "jcode_opus_minus_opencode_opus_best_score": 1.1788,
    "jcode_opus_efficiency_factor_vs_opencode_opus": 2.2639,
    "historical_jcode_opus_total_minutes": 1423.0,
    "opencode_opus_successful_total_minutes": 169.478,
    "opencode_opus_observed_minutes_including_preemption": 253.717
  },
  "infrastructure_events": [
    {
      "task": "float-print",
      "event": "modal_preemption",
      "full_gate_passed": true,
      "full_gate_score": 7.2954,
      "best_score": 7.2973,
      "successful_retry_used_in_main_table": true,
      "observed_opus_minutes_including_preemption": 253.717
    }
  ],
  "sources": {
    "repository": "https://github.com/1jehuang/jcode-bench",
    "result_commit": "3043c79",
    "opus_results": "https://github.com/1jehuang/jcode-bench/blob/3043c79/modal/runs/2026-07-18-opencode-opus48-results.json",
    "preemption_audit": "https://github.com/1jehuang/jcode-bench/blob/3043c79/modal/runs/2026-07-18-opencode-opus48-float-print-preemption.json"
  },
  "caveats": [
    "Opus has one successful run per task, so run variance is unmeasured.",
    "Modal preempted the first Opus float-print container after it passed a 7.2954 full gate; the same FunctionCall automatically restarted from scratch. Main aggregate scores use the successful retry, while the preempted best is reported separately.",
    "Successful agent duration excludes the preempted Opus float-print attempt. Observed time including preemption is an elapsed approximation from container timestamps.",
    "Historical Jcode Opus runs used much longer, unmatched budgets and older harness/Jcode versions; treat that context as directional, not a controlled agent comparison."
  ]
}