{
  "source": "https://arxiv.org/html/2605.27922v2",
  "table": 2,
  "revision": "2026-10-15",
  "task_set": "general",
  "task_count": 274,
  "metric": "mean score out of 100, not percentage of fully completed tasks",
  "runs_per_task_model_harness": 1,
  "harnesses": ["Claude Code", "Codex", "OpenCode", "OpenClaw", "Hermes", "Goose", "Nanobot"],
  "models": ["GPT-5.6 Sol", "GPT-5.6 Terra", "GPT-5.6 Luna", "GLM-5.3", "GLM-5.3-Flash", "DeepSeek-V4-Pro", "DeepSeek-V4-Flash", "Kimi-K3", "Gemma-4-31B", "Qwen3.8-27B"],
  "scores": [
    [71.6,69.8,71.0,73.1,71.2,71.5,66.5],
    [68.5,66.8,67.1,67.9,68.4,69.6,64.2],
    [69.5,67.3,66.9,67.8,68.0,68.3,64.5],
    [68.7,68.6,70.3,67.4,66.3,68.0,61.8],
    [66.3,64.8,69.5,69.0,67.4,69.9,67.2],
    [68.7,69.5,69.0,64.3,67.3,66.0,64.9],
    [64.8,68.3,66.2,66.8,67.4,65.9,60.9],
    [60.0,65.1,69.5,67.3,68.2,67.0,65.2],
    [59.0,58.9,59.3,58.8,59.6,52.7,52.2],
    [61.7,62.4,65.1,58.3,61.8,25.9,34.3]
  ]
}