真实软件工程实现与修复任务。
Raw Evidence
编码 Agent 原始证据
完整展开当前 52 组标准化公开结果。先筛选，再检查三项原始基准、成本、时间、步骤和 Token 组成。
来源Artificial Analysis Coding Agents
指数版本v1.2
记录数52
快照2026-08-06
ALL 52 RESULTS
完整数据表
移动端保留横向滚动；首列固定，避免模型身份在检查指标时丢失。
| # | Agent / 模型 | 模型机构 | 综合指数 | DeepSWE | Terminal-Bench v2 | SWE-Atlas-QnA | 成本 / 任务 | 执行时间 | 平均步骤 | 输入 Token | 缓存 Token | 输出 Token | 总 Token | 缓存命中率 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude CodeOpus 5 (xhigh) | Anthropic | 66.7 | 60.5 | 84.9 | 54.8 | US$8.23 | 23.6 分钟 | 153 | 10,876,804 | 10,707,864 | 72,794 | 21,823,530 | 97.5% |
| 2 | CodexGPT-5.6 Sol (max) | OpenAI | 66.6 | 68.7 | 87.7 | 43.3 | US$7.08 | 10.2 分钟 | 114 | 6,812,390 | 6,360,943 | 54,860 | 13,228,192 | 89.9% |
| 3 | Claude CodeFable 5 (max) | Anthropic | 65.8 | 66.1 | 82.5 | 48.9 | US$11.71 | 23.4 分钟 | 138 | 6,958,612 | 6,798,600 | 73,567 | 13,982,074 | 96.2% |
| 4 | Claude CodeOpus 5 (max) | Anthropic | 65.5 | 63.1 | 84.5 | 48.9 | US$8.95 | 23.7 分钟 | 166 | 11,907,546 | 11,732,777 | 80,473 | 23,891,138 | 97.5% |
| 5 | CodexGPT-5.6 Sol (xhigh) | OpenAI | 65.1 | 67.0 | 86.1 | 42.2 | US$5.24 | 7.4 分钟 | 96 | 5,120,783 | 4,781,615 | 38,288 | 9,940,686 | 90.6% |
| 6 | Grok BuildGrok 4.5 (high) | xAI | 64.4 | 59.9 | 85.3 | 48.1 | US$2.59 | 16.5 分钟 | 61 | 1,820,320 | 1,717,419 | 39,966 | 3,577,706 | 92.5% |
| 7 | CodexGPT-5.6 Sol (high) | OpenAI | 64.1 | 64.9 | 82.5 | 44.9 | US$4.14 | 6.3 分钟 | 86 | 4,163,502 | 3,893,166 | 28,199 | 8,084,867 | 90.2% |
| 8 | Claude CodeOpus 5 (high) | Anthropic | 63.4 | 60.8 | 80.2 | 49.2 | US$3.80 | 13.4 分钟 | 93 | 4,832,263 | 4,746,535 | 35,447 | 9,699,126 | 97.1% |
| 9 | CodexGPT-5.6 Terra (max) | OpenAI | 62.3 | 67.0 | 84.1 | 35.8 | US$2.21 | 8.4 分钟 | 97 | 4,843,755 | 4,561,390 | 60,732 | 9,465,877 | 91.4% |
| 10 | Claude CodeOpus 5 (medium) | Anthropic | 61.9 | 62.8 | 78.6 | 44.4 | US$3.14 | 12.2 分钟 | 83 | 3,953,917 | 3,880,763 | 29,632 | 7,936,908 | 97.0% |
| 11 | CodexGPT-5.5 (xhigh) | OpenAI | 61.5 | 64.3 | 84.1 | 36.0 | US$5.07 | 10.1 分钟 | 106 | 6,203,311 | 6,018,972 | 37,947 | 12,260,230 | 93.9% |
| 12 | Kimi Code CLIKimi K3 | Moonshot AI | 61.3 | 63.7 | 83.7 | 36.6 | US$3.18 | 23.8 分钟 | 125 | 5,400,248 | 5,130,236 | 88,806 | 10,619,290 | 95.0% |
| 13 | CodexGPT-5.6 Sol (medium) | OpenAI | 60.6 | 64.0 | 77.8 | 40.1 | US$2.99 | 5.2 分钟 | 72 | 3,001,126 | 2,797,277 | 19,115 | 5,817,519 | 89.5% |
| 14 | Claude CodeOpus 4.8 (max) | Anthropic | 60.5 | 55.8 | 79.4 | 46.5 | US$7.70 | 23.1 分钟 | 166 | 8,917,632 | 8,730,314 | 88,631 | 17,913,878 | 96.5% |
| 15 | CodexGPT-5.6 Luna (max) | OpenAI | 58.7 | 63.4 | 79.8 | 32.8 | US$0.31 | 8.0 分钟 | 115 | 7,913,897 | 7,485,097 | 64,941 | 15,463,936 | 91.4% |
| 16 | Claude CodeOpus 4.8 (xhigh) | Anthropic | 58.5 | 51.3 | 81.3 | 42.7 | US$5.65 | 17.6 分钟 | 136 | 6,805,832 | 6,683,709 | 61,811 | 13,671,704 | 96.9% |
| 17 | CodexGPT-5.6 Terra (xhigh) | OpenAI | 57.1 | 58.4 | 80.6 | 32.3 | US$1.52 | 6.9 分钟 | 75 | 3,313,555 | 3,107,178 | 40,160 | 6,460,892 | 90.7% |
| 18 | Claude CodeOpus 5 (low) | Anthropic | 56.8 | 56.9 | 74.2 | 39.2 | US$2.18 | 9.5 分钟 | 64 | 2,568,845 | 2,510,220 | 22,281 | 5,159,612 | 96.5% |
| 19 | Claude CodeOpus 4.8 (high) | Anthropic | 56.8 | 51.6 | 79.8 | 39.0 | US$3.72 | 12.4 分钟 | 104 | 4,543,634 | 4,462,984 | 39,605 | 9,125,788 | 96.9% |
| 20 | CodexGPT-5.6 Terra (high) | OpenAI | 55.8 | 60.5 | 76.0 | 30.9 | US$1.27 | 6.2 分钟 | 67 | 2,843,915 | 2,664,435 | 31,662 | 5,540,012 | 90.3% |
| 21 | CodexGPT-5.6 Luna (xhigh) | OpenAI | 54.7 | 56.6 | 76.2 | 31.2 | US$0.25 | 6.6 分钟 | 96 | 6,321,539 | 5,942,041 | 47,114 | 12,310,694 | 90.4% |
| 22 | CodexGPT-5.5 (medium) | OpenAI | 54.4 | 56.6 | 75.8 | 30.6 | US$2.75 | 6.4 分钟 | 78 | 3,525,549 | 3,419,805 | 17,196 | 6,962,550 | 94.8% |
| 23 | CodexGPT-5.6 Sol (low) | OpenAI | 53.6 | 53.4 | 73.0 | 34.4 | US$1.72 | 3.7 分钟 | 54 | 1,641,547 | 1,513,072 | 10,620 | 3,165,239 | 88.4% |
| 24 | Claude CodeOpus 4.8 (medium) | Anthropic | 53.6 | 49.3 | 75.4 | 36.0 | US$3.26 | 12.4 分钟 | 93 | 3,870,142 | 3,787,948 | 34,207 | 7,769,574 | 96.5% |
| 25 | OpencodeMuse Spark 1.1 (xhigh) | Meta | 53.5 | 54.3 | 73.0 | 33.3 | US$1.43 | 12.6 分钟 | 55 | 6,261,271 | 5,948,208 | 34,042 | 12,243,521 | 95.0% |
| 26 | CodexGPT-5.6 Luna (high) | OpenAI | 51.4 | 53.4 | 71.8 | 29.0 | US$0.19 | 5.7 分钟 | 84 | 4,888,008 | 4,581,094 | 32,273 | 9,501,375 | 89.9% |
| 27 | Claude CodeOpus 4.7 (max) | Anthropic | 50.3 | 40.1 | 73.8 | 37.1 | US$5.63 | 15.7 分钟 | 107 | 7,948,863 | 7,818,313 | 46,068 | 15,941,371 | 96.9% |
| 28 | OpencodeOpus 4.7 (medium) | Anthropic | 50.0 | 39.5 | 75.4 | 34.9 | US$2.93 | 12.2 分钟 | 54 | 3,734,060 | 3,734,001 | 25,751 | 7,561,040 | 96.1% |
| 29 | CodexGPT-5.6 Terra (medium) | OpenAI | 47.8 | 45.7 | 69.4 | 28.2 | US$0.72 | 4.3 分钟 | 51 | 1,614,396 | 1,499,910 | 16,002 | 3,130,308 | 89.3% |
| 30 | Claude CodeOpus 4.8 (low) | Anthropic | 47.4 | 41.0 | 73.0 | 28.2 | US$2.15 | 8.4 分钟 | 67 | 2,571,741 | 2,520,479 | 22,820 | 5,165,422 | 96.7% |
| 31 | Claude CodeOpus 4.6 (medium) | Anthropic | 46.5 | — | 70.6 | 22.3 | US$1.28 | 8.0 分钟 | 34 | 2,234,510 | 2,110,648 | 19,008 | 4,476,378 | 93.6% |
| 32 | Cursor CLIGPT-5.5 (medium) | OpenAI | 46.1 | 37.2 | 73.4 | 27.7 | US$2.01 | 6.6 分钟 | 78 | 2,061,667 | 1,916,592 | 11,044 | 3,989,303 | 89.0% |
| 33 | Cursor CLIOpus 4.7 (medium) | Anthropic | 45.4 | 31.6 | 70.6 | 33.9 | US$2.68 | 13.6 分钟 | 86 | 2,819,480 | 2,745,220 | 17,782 | 5,656,580 | 95.9% |
| 34 | CodexGPT-5.6 Sol (none) | OpenAI | 43.4 | 35.4 | 60.7 | 34.1 | US$1.40 | 3.4 分钟 | 55 | 1,737,377 | 1,670,776 | 7,663 | 3,415,816 | 89.0% |
| 35 | Claude CodeGLM-5.2 | Z.ai | 43.2 | 28.6 | 71.9 | 29.0 | US$6.51 | 25.1 分钟 | 127 | 5,424,051 | 1,052,877 | 40,310 | 6,517,238 | 34.1% |
| 36 | CodexGPT-5.6 Luna (medium) | OpenAI | 42.4 | 36.6 | 63.5 | 27.2 | US$0.09 | 3.4 分钟 | 58 | 2,273,504 | 2,100,885 | 15,198 | 4,389,586 | 87.9% |
| 37 | Claude CodeOpus 4.7 (medium) | Anthropic | 40.5 | 27.4 | 71.4 | 22.6 | US$1.68 | 6.3 分钟 | 42 | 2,279,493 | 2,224,627 | 16,673 | 4,574,241 | 95.8% |
| 38 | CodexGPT-5.4 (medium) | OpenAI | 39.1 | 25.0 | 69.8 | 22.3 | US$2.42 | 7.1 分钟 | 73 | 3,011,113 | 2,895,710 | 18,149 | 5,924,972 | 92.9% |
| 39 | Cursor CLIComposer 2.5 | Cursor | 38.2 | 15.9 | 67.5 | 31.2 | US$0.08 | 9.5 分钟 | 117 | 1,815,576 | 1,747,519 | 17,531 | 3,577,610 | 93.9% |
| 40 | Cursor CLIComposer 2.5 Fast | Cursor | 38.2 | 15.9 | 67.5 | 31.2 | US$0.55 | 6.8 分钟 | 117 | 2,148,692 | 2,074,061 | 19,638 | 4,238,247 | 93.6% |
| 41 | Claude CodeSonnet 4.6 (medium) | Anthropic | 37.6 | 28.9 | 64.3 | 19.6 | US$2.01 | 13.5 分钟 | 67 | 4,224,651 | 4,102,518 | 38,508 | 8,481,319 | 94.9% |
| 42 | Cursor CLIGPT-5.4 (medium) | OpenAI | 37.1 | 16.7 | 66.3 | 28.2 | US$1.55 | 8.1 分钟 | 26 | 2,129,535 | 1,872,655 | 15,483 | 4,017,673 | 85.4% |
| 43 | CodexGPT-5.6 Terra (low) | OpenAI | 36.7 | 29.8 | 57.5 | 22.8 | US$0.39 | 2.8 分钟 | 37 | 797,366 | 724,394 | 8,053 | 1,529,813 | 87.7% |
| 44 | Claude CodeGLM-5.1 | Z.ai | 36.1 | 18.6 | 65.1 | 24.7 | US$4.33 | 19.6 分钟 | 174 | 13,212,122 | 12,620,806 | 50,447 | 25,883,374 | 87.8% |
| 45 | Claude CodeQwen3.7 Plus (thinking) | Alibaba Cloud | 36.0 | 19.2 | 65.1 | 23.7 | US$6.23 | 10.6 分钟 | 146 | 4,668,456 | 3,995,341 | 32,127 | 8,695,924 | 81.0% |
| 46 | Claude CodeKimi K2.6 | Moonshot AI | 32.6 | 16.5 | 65.5 | 15.9 | US$1.19 | 41.0 分钟 | 131 | 5,786,199 | 5,631,883 | 36,275 | 11,454,356 | 95.8% |
| 47 | Claude CodeDeepSeek V4 Pro (high) | DeepSeek | 31.4 | 8.6 | 65.9 | 19.9 | US$0.27 | 17.9 分钟 | 127 | 5,143,082 | 4,638,910 | 41,153 | 9,823,145 | 83.3% |
| 48 | Gemini CLIGemini 3.1 Pro (high) | 30.3 | 14.2 | 68.3 | 8.6 | US$2.00 | 10.8 分钟 | 31 | 2,448,999 | 2,236,947 | 18,426 | 4,704,372 | 87.0% | |
| 49 | Cursor CLIComposer 2 | Cursor | 27.5 | 0.0 | 64.7 | 17.7 | US$0.04 | 8.6 分钟 | 28 | 1,476,106 | 1,419,615 | 14,007 | 2,954,431 | 92.6% |
| 50 | CodexGPT-5.6 Luna (low) | OpenAI | 25.1 | 10.3 | 49.6 | 15.3 | US$0.04 | 1.9 分钟 | 35 | 798,108 | 699,715 | 6,699 | 1,504,521 | 84.2% |
| 51 | CodexGPT-5.6 Terra (none) | OpenAI | 23.7 | 13.3 | 39.3 | 18.5 | US$0.30 | 1.8 分钟 | 34 | 587,769 | 523,111 | 5,171 | 1,116,050 | 85.3% |
| 52 | CodexGPT-5.6 Luna (none) | OpenAI | 20.4 | 6.5 | 37.3 | 17.5 | US$0.07 | 2.5 分钟 | 56 | 1,845,654 | 1,712,567 | 7,936 | 3,566,157 | 87.6% |
“—”表示来源没有公开可核验数据。Token 为每项任务的平均用量；厂商受控实验不插入这张标准化数据表。
多步骤终端与工具调用工作流。
代码库理解与技术问答。