Benchmark
SWE-rebench 2025-05
Coding benchmarks compare models on programming tasks. Scores are percentages.
- Models measured
- 18
- Scores recorded
- 23
- Sources
- 1
- IQ weight
- 1.42
Score distribution
- 26.9%
- Mean
- 24%
- Median
- 14.2%
- Std dev
- 8.5%
- 10th percentile
- 42.5%
- 90th percentile
- 0.5%
- Min
This benchmark has a weight of 1.42 for chat models, measured over 18 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4tools· SWE-rebench | 56.5% |
| 2 | OpenAI o3tools· SWE-rebench | 49% |
| 3 | GPT 4.1text· SWE-rebench | 42.5% |
| 4 | GPT 4.1tools· SWE-rebench | 41.5% |
| 5 | Gemini 2.5 Flash Preview 05-20 (no think)text, no-thinking· SWE-rebench | 41% |
| 6 | DeepSeek V3tools· SWE-rebench | 38% |
| 7 | Claude 3.5 Sonnettools· SWE-rebench | 37% |
| 8 | GPT-4.1 minitext· SWE-rebench | 35.5% |
| 9 | DeepSeek V3text· SWE-rebench | 34% |
| 10 | Gemini 2.5 Flash Preview 05-20 (no think)tools, no-thinking· SWE-rebench | 31% |
| 11 | Qwen 3 235B A22Btext, no-thinking· SWE-rebench | 27.5% |
| 12 | Qwen3 32Btext, thinking· SWE-rebench | 24% |
| 13 | Qwen 3 235B A22Btext, thinking· SWE-rebench | 23.5% |
| 14 | Devstral Small 2505text, 2505· SWE-rebench | 23% |
| 15 | Qwen3 32Btext, no-thinking· SWE-rebench | 22% |
| 16 | Llama 3.3 70B Instructtext· SWE-rebench | 21% |
| 17 | Llama 4 Mavericktext· SWE-rebench | 16.5% |
| 18 | Qwen2.5 72B Instructtext· SWE-rebench | 16% |
| 19 | Gemini 2.0 Flashtext· SWE-rebench | 12.5% |
| 20 | Gemma 3 27B ITtext· SWE-rebench | 9.5% |
| 21 | Llama 4 Scouttext· SWE-rebench | 8.5% |
| 22 | Qwen 2.5 Coder 32B Instructtext· SWE-rebench | 7.5% |
| 23 | GPT-4.1 nanotext· SWE-rebench | 0.5% |