编程智能体排行榜
真实软件工程任务中的智能体表现、每任务 API 成本与执行时间
Coding Agent Index v1.5快照日期: 2026-10-22
排序
| # | 智能体 / 模型 | 指数 | 每任务成本 | 每任务耗时 |
|---|---|---|---|---|
| 1 | Claude Code | 68.4%最佳 | US$14.19 | 1.5h |
| 2 | Devin Fusion CLI | 68.2% | US$9.02 | 55m |
| 3 | Muse Code Muse Wonder Full (max) | 68.0% | — | 28.8m |
| 4 | Claude Code | 66.0% | US$13.04 | 1.1h |
| 5 | Antigravity CLI | 63.8% | US$5.84 | 34.5m |
| 6 | Muse Code | 63.7% | US$8.04 | 22.4m |
| 7 | Claude Code | 62.2% | US$12.39 | 34.8m |
| 8 | Claude Code | 61.7% | US$11.42 | 1h |
| 9 | Codex | 61.6% | US$7.50 | 29.4m |
| 10 | Devin Fusion CLI | 61.5% | US$8.00 | 51.1m |
| 11 | Codex | 60.1% | US$1.55 | 24.4m |
| 12 | GitHub Copilot Copilot Fusion (max) | 59.9% | US$5.59 | 36.1m |
| 13 | Claude Code | 59.3% | US$13.30 | 1.3h |
| 14 | Muse Code Muse Wonder Full (Max) | 58.9% | — | 34.9m |
| 15 | Claude Code | 57.6% | US$11.53 | 35.3m |
| 16 | Grok Build | 56.3% | US$8.82 | 39.2m |
| 17 | Codex | 54.7% | US$7.04 | 24.8m |
| 18 | Codex | 54.1% | US$1.50 | 23.7m |
| 19 | Opencode | 53.6% | US$4.37 | 48.1m |
| 20 | Antigravity CLI | 53.2% | US$11.73 | 34.7m |
| 21 | Muse Code | 52.0% | US$8.05 | 24.5m |
| 22 | Kimi Code CLI | 51.9% | US$5.05 | 1h |
| 23 | Opencode | 48.9% | US$5.23 | 56.2m |
| 24 | Mistral Vibe | 48.3% | US$1.26 | 48.4m |
| 25 | Codex | 48.2% | US$0.69 | 16.4m |
| 26 | Grok Build | 46.6% | US$9.48 | 35.8m |
| 27 | DeepSeek Harness | 46.3% | US$0.30 | 21.4m |
| 28 | Kimi Code CLI | 41.4% | US$5.17 | 1h |
| 29 | Muse Code | 41.2% | US$3.33 | 25.2m |
| 30 | Claude Code | 36.8% | US$2.80 | 1.5h |
| 31 | Mirror | 27.7% | — | 42.4m |
| 32 | Mirror | 20.5% | US$0.32 | 38.7m |
指数如何构成
Artificial Analysis 让每个智能体在三个基准上各跑三遍取 pass@1：DeepSWE v1.1（113 个软件工程任务）、Terminal-Bench 4.0（66 个终端任务）与 SWE-Atlas-QnA（124 个技术问答）。综合指数为三项等权平均。成本按供应商 token 报价 × 实测 token 用量计算，不含订阅套餐。