{
  "_comment": "GENERATED by scripts/gen_leaderboard.py - do not edit by hand. Each row's `id` maps to assets/details/<id>.json and detail.html?id=<id>. costPerRun is USD for one full pass over the task suite (total cost / attempts per task) and also feeds the score-vs-cost chart.",
  "logos": {
    "openai": {
      "src": "assets/models/openai.svg",
      "adaptive": true
    },
    "claude": {
      "src": "assets/models/claude.svg",
      "adaptive": false
    },
    "zai": {
      "src": "assets/models/zai.svg",
      "adaptive": true
    },
    "gemini": {
      "src": "assets/models/gemini.svg",
      "adaptive": false
    },
    "deepseek": {
      "src": "assets/models/deepseek.svg",
      "adaptive": false
    },
    "minimax": {
      "src": "assets/models/minimax.svg",
      "adaptive": false
    },
    "qwen": {
      "src": "assets/models/qwen-color.svg",
      "adaptive": false
    },
    "kimi": {
      "src": "assets/models/kimi-ai.svg",
      "adaptive": false
    }
  },
  "rows": [
    {
      "id": "claude_code_claude_opus_4_8_max",
      "agent": "Claude Code",
      "model": "claude-opus-4.8",
      "logo": "claude",
      "thinking": "max",
      "successRate": 0.658,
      "successError": 0.007,
      "pass1": 58.8,
      "pass5": 64.2,
      "all5": 51.7,
      "costPerRun": 173.61
    },
    {
      "id": "codex_gpt_5_5_xhigh",
      "agent": "Codex",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "xhigh",
      "successRate": 0.647,
      "successError": 0.007,
      "pass1": 57.7,
      "pass5": 68.3,
      "all5": 42.5,
      "costPerRun": 138.4
    },
    {
      "id": "codex_gpt_5_5_high",
      "agent": "Codex",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "high",
      "successRate": 0.642,
      "successError": 0.007,
      "pass1": 57.2,
      "pass5": 66.7,
      "all5": 46.7,
      "costPerRun": 104.87
    },
    {
      "id": "openhands_sdk_claude_opus_4_8_max",
      "agent": "OpenHands",
      "model": "claude-opus-4.8",
      "logo": "claude",
      "thinking": "max",
      "successRate": 0.634,
      "successError": 0.006,
      "pass1": 57.3,
      "pass5": 67.5,
      "all5": 45.0,
      "costPerRun": 196.3
    },
    {
      "id": "mini_swe_agent_gpt_5_5_xhigh",
      "agent": "Mini-SWE-Agent",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "xhigh",
      "successRate": 0.624,
      "successError": 0.008,
      "pass1": 54.2,
      "pass5": 67.5,
      "all5": 40.0,
      "costPerRun": 186.38
    },
    {
      "id": "openhands_sdk_gpt_5_5_xhigh",
      "agent": "OpenHands",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "xhigh",
      "successRate": 0.614,
      "successError": 0.01,
      "pass1": 54.0,
      "pass5": 65.0,
      "all5": 38.3,
      "costPerRun": 177.15
    },
    {
      "id": "terminus_2_gpt_5_5_xhigh",
      "agent": "Terminus-2",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "xhigh",
      "successRate": 0.601,
      "successError": 0.006,
      "pass1": 52.3,
      "pass5": 64.2,
      "all5": 31.7,
      "costPerRun": 116.06
    },
    {
      "id": "terminus_2_claude_opus_4_8_max",
      "agent": "Terminus-2",
      "model": "claude-opus-4.8",
      "logo": "claude",
      "thinking": "max",
      "successRate": 0.597,
      "successError": 0.01,
      "pass1": 53.8,
      "pass5": 62.5,
      "all5": 42.5,
      "costPerRun": 183.74
    },
    {
      "id": "codex_gpt_5_5_medium",
      "agent": "Codex",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "medium",
      "successRate": 0.588,
      "successError": 0.01,
      "pass1": 51.2,
      "pass5": 62.5,
      "all5": 34.2,
      "costPerRun": 81.79
    },
    {
      "id": "terminus_2_claude_opus_4_7_max",
      "agent": "Terminus-2",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "max",
      "successRate": 0.58,
      "successError": 0.008,
      "pass1": 51.0,
      "pass5": 64.2,
      "all5": 39.2,
      "costPerRun": 131.26
    },
    {
      "id": "terminus_2_gpt_5_5_high",
      "agent": "Terminus-2",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "high",
      "successRate": 0.578,
      "successError": 0.017,
      "pass1": 49.8,
      "pass5": 63.3,
      "all5": 32.5,
      "costPerRun": 93.75
    },
    {
      "id": "mini_swe_agent_claude_opus_4_8_max",
      "agent": "Mini-SWE-Agent",
      "model": "claude-opus-4.8",
      "logo": "claude",
      "thinking": "max",
      "successRate": 0.574,
      "successError": 0.006,
      "pass1": 50.2,
      "pass5": 64.2,
      "all5": 34.2,
      "costPerRun": 57.14
    },
    {
      "id": "terminus_2_claude_opus_4_7_xhigh",
      "agent": "Terminus-2",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "xhigh",
      "successRate": 0.554,
      "successError": 0.008,
      "pass1": 48.0,
      "pass5": 59.2,
      "all5": 37.5,
      "costPerRun": 72.8
    },
    {
      "id": "terminus_2_gpt_5_5_medium",
      "agent": "Terminus-2",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "medium",
      "successRate": 0.515,
      "successError": 0.013,
      "pass1": 42.8,
      "pass5": 62.5,
      "all5": 21.7,
      "costPerRun": 54.0
    },
    {
      "id": "claude_code_claude_opus_4_7_xhigh",
      "agent": "Claude Code",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "xhigh",
      "successRate": 0.503,
      "successError": 0.01,
      "pass1": 43.0,
      "pass5": 57.5,
      "all5": 29.2,
      "costPerRun": 113.95
    },
    {
      "id": "claude_code_claude_opus_4_7_max",
      "agent": "Claude Code",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "max",
      "successRate": 0.501,
      "successError": 0.006,
      "pass1": 42.8,
      "pass5": 55.8,
      "all5": 30.8,
      "costPerRun": 144.18
    },
    {
      "id": "terminus_2_claude_opus_4_7_high",
      "agent": "Terminus-2",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "high",
      "successRate": 0.499,
      "successError": 0.012,
      "pass1": 42.5,
      "pass5": 58.3,
      "all5": 25.0,
      "costPerRun": 60.96
    },
    {
      "id": "terminus_2_claude_opus_4_7_none",
      "agent": "Terminus-2",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "none",
      "successRate": 0.497,
      "successError": 0.011,
      "pass1": 41.7,
      "pass5": 58.3,
      "all5": 27.5,
      "costPerRun": 54.27
    },
    {
      "id": "terminus_2_gemini_3_1_pro_preview_high",
      "agent": "Terminus-2",
      "model": "gemini-3.1-pro-preview",
      "logo": "gemini",
      "thinking": "high",
      "successRate": 0.493,
      "successError": 0.018,
      "pass1": 41.2,
      "pass5": 57.5,
      "all5": 24.2,
      "costPerRun": 68.57
    },
    {
      "id": "mini_swe_agent_gemini_3_1_pro_preview_none",
      "agent": "Mini-SWE-Agent",
      "model": "gemini-3.1-pro-preview",
      "logo": "gemini",
      "thinking": "none",
      "successRate": 0.485,
      "successError": 0.015,
      "pass1": 40.0,
      "pass5": 57.5,
      "all5": 20.8,
      "costPerRun": 199.22
    },
    {
      "id": "terminus_2_z_ai_glm_5_1_xhigh",
      "agent": "Terminus-2",
      "model": "glm-5.1",
      "logo": "zai",
      "thinking": "xhigh",
      "successRate": 0.481,
      "successError": 0.013,
      "pass1": 40.3,
      "pass5": 59.2,
      "all5": 20.8,
      "costPerRun": 23.49
    },
    {
      "id": "claude_code_claude_opus_4_7_high",
      "agent": "Claude Code",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "high",
      "successRate": 0.474,
      "successError": 0.004,
      "pass1": 40.7,
      "pass5": 54.2,
      "all5": 27.5,
      "costPerRun": 88.0
    },
    {
      "id": "terminus_2_minimax_minimax_m3_xhigh",
      "agent": "Terminus-2",
      "model": "minimax-m3",
      "logo": "minimax",
      "thinking": "xhigh",
      "successRate": 0.47,
      "successError": 0.013,
      "pass1": 41.2,
      "pass5": 59.2,
      "all5": 22.5,
      "costPerRun": 12.03
    },
    {
      "id": "codex_gpt_5_5_low",
      "agent": "Codex",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "low",
      "successRate": 0.467,
      "successError": 0.011,
      "pass1": 39.0,
      "pass5": 61.7,
      "all5": 20.0,
      "costPerRun": 41.62
    },
    {
      "id": "terminus_2_deepseek_deepseek_v4_pro_xhigh",
      "agent": "Terminus-2",
      "model": "deepseek-v4-pro",
      "logo": "deepseek",
      "thinking": "xhigh",
      "successRate": 0.462,
      "successError": 0.008,
      "pass1": 38.0,
      "pass5": 57.5,
      "all5": 18.3,
      "costPerRun": 47.14
    },
    {
      "id": "terminus_2_claude_opus_4_7_medium",
      "agent": "Terminus-2",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "medium",
      "successRate": 0.457,
      "successError": 0.007,
      "pass1": 37.8,
      "pass5": 51.7,
      "all5": 23.3,
      "costPerRun": 40.12
    },
    {
      "id": "claude_code_claude_opus_4_7_none",
      "agent": "Claude Code",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "none",
      "successRate": 0.45,
      "successError": 0.011,
      "pass1": 37.7,
      "pass5": 50.8,
      "all5": 24.2,
      "costPerRun": 94.87
    },
    {
      "id": "terminus_2_qwen_qwen3_7_max_xhigh",
      "agent": "Terminus-2",
      "model": "qwen3.7-max",
      "logo": "qwen",
      "thinking": "xhigh",
      "successRate": 0.449,
      "successError": 0.007,
      "pass1": 37.7,
      "pass5": 57.5,
      "all5": 21.7,
      "costPerRun": 19.04
    },
    {
      "id": "openhands_sdk_gemini_3_1_pro_preview_none",
      "agent": "OpenHands",
      "model": "gemini-3.1-pro-preview",
      "logo": "gemini",
      "thinking": "none",
      "successRate": 0.441,
      "successError": 0.015,
      "pass1": 35.8,
      "pass5": 56.7,
      "all5": 19.2,
      "costPerRun": 303.84
    },
    {
      "id": "terminus_2_claude_sonnet_4_6_max",
      "agent": "Terminus-2",
      "model": "claude-sonnet-4.6",
      "logo": "claude",
      "thinking": "max",
      "successRate": 0.428,
      "successError": 0.003,
      "pass1": 34.8,
      "pass5": 49.2,
      "all5": 20.0,
      "costPerRun": 37.36
    },
    {
      "id": "terminus_2_moonshotai_kimi_k2_6_xhigh",
      "agent": "Terminus-2",
      "model": "kimi-k2.6",
      "logo": "kimi",
      "thinking": "xhigh",
      "successRate": 0.428,
      "successError": 0.018,
      "pass1": 35.3,
      "pass5": 55.8,
      "all5": 18.3,
      "costPerRun": 22.71
    },
    {
      "id": "terminus_2_gpt_5_5_low",
      "agent": "Terminus-2",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "low",
      "successRate": 0.424,
      "successError": 0.017,
      "pass1": 33.5,
      "pass5": 51.7,
      "all5": 13.3,
      "costPerRun": 32.71
    },
    {
      "id": "claude_code_claude_opus_4_7_medium",
      "agent": "Claude Code",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "medium",
      "successRate": 0.417,
      "successError": 0.006,
      "pass1": 34.8,
      "pass5": 50.8,
      "all5": 19.2,
      "costPerRun": 64.76
    },
    {
      "id": "terminus_2_claude_opus_4_7_low",
      "agent": "Terminus-2",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "low",
      "successRate": 0.412,
      "successError": 0.01,
      "pass1": 32.7,
      "pass5": 45.8,
      "all5": 21.7,
      "costPerRun": 31.79
    },
    {
      "id": "terminus_2_gpt_5_5_none",
      "agent": "Terminus-2",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "none",
      "successRate": 0.365,
      "successError": 0.008,
      "pass1": 28.2,
      "pass5": 45.0,
      "all5": 14.2,
      "costPerRun": 29.57
    },
    {
      "id": "claude_code_claude_opus_4_7_low",
      "agent": "Claude Code",
      "model": "claude-opus-4.7",
      "logo": "claude",
      "thinking": "low",
      "successRate": 0.353,
      "successError": 0.003,
      "pass1": 28.3,
      "pass5": 40.8,
      "all5": 16.7,
      "costPerRun": 55.4
    },
    {
      "id": "codex_gpt_5_5_none",
      "agent": "Codex",
      "model": "gpt-5.5",
      "logo": "openai",
      "thinking": "none",
      "successRate": 0.35,
      "successError": 0.009,
      "pass1": 26.2,
      "pass5": 49.2,
      "all5": 8.3,
      "costPerRun": 30.6
    },
    {
      "id": "terminus_2_gpt_5_4_mini_xhigh",
      "agent": "Terminus-2",
      "model": "gpt-5.4-mini",
      "logo": "openai",
      "thinking": "xhigh",
      "successRate": 0.272,
      "successError": 0.014,
      "pass1": 20.0,
      "pass5": 41.7,
      "all5": 6.7,
      "costPerRun": 148.83
    },
    {
      "id": "terminus_2_claude_haiku_4_5_none",
      "agent": "Terminus-2",
      "model": "claude-haiku-4.5",
      "logo": "claude",
      "thinking": "none",
      "successRate": 0.239,
      "successError": 0.015,
      "pass1": 15.7,
      "pass5": 30.8,
      "all5": 3.3,
      "costPerRun": 52.15
    }
  ]
}