AI 评测与排行榜，分别告诉你什么
分清评测基准、一次评测运行与排行榜，用它们回答各自适合的问题。
Model reports
SWE-bench Verified
Scaffold, tool permissions and time limit are part of the result; vendors run their own harnesses.
评测信息与证据
- 来源
- Model reports
- 发布者
- 未收录
- 评测发布日期
- 2024-08-13
- 采集日期
- 未收录
- 模态
- 未收录
本站保留来源报告的结果，不代表独立复测。请先核对任务版本、评测协议、分数单位和报告日期；不同运行条件不能直接合并为统一排名。
原始资料
报告成绩 16
按来源中的观察顺序展示，顺序不代表排名。日期保留原始精度。
| 模型 | 数值 | 单位 | 协议 | 报告日期与口径 | 来源 |
|---|---|---|---|---|---|
| Claude 3.7 Sonnet | 70.3 | percent | parallel attempts plus internal ranker, 489-task infrastructure-compatible subset | 2025-02-24 · document_publication | Claude 3.7 Sonnet |
| Claude 3.7 Sonnet | 63.7 | percent | vanilla pass@1, 489-task infrastructure-compatible subset | 2025-02-24 · document_publication | Claude 3.7 Sonnet |
| Claude Opus 4 | 72.5 | percent | no extended thinking | 2025-05-22 · document_publication | Claude Opus 4 and Sonnet 4 |
| Claude Sonnet 4 | 72.7 | percent | no extended thinking | 2025-05-22 · document_publication | Claude Opus 4 and Sonnet 4 |
| Claude Opus 5 | 96.0 | percent | adaptive thinking, max effort, averaged over 5 trials | 2026-07-24 · document_publication | Claude Opus 5 |
| DeepSeek-R1 | 49.2 | percent | Resolved, vendor harness | 2025-01-22 · document_publication | DeepSeek-R1 |
| DeepSeek-V3 | 42.0 | percent | Resolved, vendor harness | 2024-12-27 · document_publication | DeepSeek-V3 |
| DeepSeek-V4-Pro | 73.6 | percent | non-think, resolved | 2026-04-22 · document_publication | DeepSeek-V4-Pro |
| DeepSeek-V4-Pro | 79.4 | percent | think high, resolved | 2026-04-22 · document_publication | DeepSeek-V4-Pro |
| DeepSeek-V4-Pro | 80.6 | percent | think max, resolved | 2026-04-22 · document_publication | DeepSeek-V4-Pro |
| DeepSeek-V4-Flash | 79.0 | percent | think max, resolved | 2026-06-25 · document_publication | DeepSeek-V4 |
| Gemini 2.5 Pro | 67.2 | percent | multiple attempts | 2025-06-17 · document_publication | Gemini 2.5 Pro |
| Gemini 2.5 Pro | 59.6 | percent | single attempt | 2025-06-17 · document_publication | Gemini 2.5 Pro |
| Gemini 3.1 Pro | 80.6 | percent | Thinking (High), Single attempt | 2026-02-19 · document_publication | Gemini 3.1 Pro |
| Qwen3.5-397B-A17B | 76.4 | percent | thinking (temp 0.6, top_p 0.95, top_k 20) | 2026-02-16 · document_publication | Qwen3.5-397B-A17B |
| GLM-5 | 77.8 | percent | OpenHands, tailored instruction prompt, temp 0.7, top_p 0.95, 200K context | 2026-02-11 · document_publication | GLM-5 |