Benchmark
SWE-rebench 2025-06
Coding benchmarks compare models on programming tasks. Scores are percentages.
- Models measured
- 17
- Scores recorded
- 21
- Sources
- 1
- IQ weight
- 1.72
Score distribution
- 22.3%
- Mean
- 18%
- Median
- 12.8%
- Std dev
- 9%
- 10th percentile
- 39%
- 90th percentile
- 0.5%
- Min
This benchmark has a weight of 1.72 for chat models, measured over 17 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 | 48.5% |
| 2 | GPT 4.1tools· SWE-rebench | 39.5% |
| 3 | OpenAI o3tools· SWE-rebench | 39% |
| 4 | DeepSeek V3text· SWE-rebench | 37% |
| 5 | GPT 4.1text· SWE-rebench | 36% |
| 6 | DeepSeek V3tools· SWE-rebench | 32.5% |
| 7 | Claude 3.5 Sonnettools· SWE-rebench | 31.5% |
| 8 | GPT-4.1 minitext· SWE-rebench | 31.5% |
| 9 | Qwen 3 235B A22Btext, no-thinking· SWE-rebench | 22% |
| 10 | Llama 4 Mavericktext· SWE-rebench | 18.5% |
| 11 | Llama 3.3 70B Instructtext· SWE-rebench | 18% |
| 12 | Qwen 3 235B A22Btext, thinking· SWE-rebench | 17% |
| 13 | Devstral Small 2505text, 2505· SWE-rebench | 16.5% |
| 14 | Qwen3 32Btext, no-thinking· SWE-rebench | 15.5% |
| 15 | Qwen2.5 72B Instructtext· SWE-rebench | 14% |
| 16 | Qwen3 32Btext, thinking· SWE-rebench | 14% |
| 17 | Llama 4 Scouttext· SWE-rebench | 13.5% |
| 18 | Gemini 2.0 Flashtext· SWE-rebench | 9.5% |
| 19 | Gemma 3 27B ITtext· SWE-rebench | 9% |
| 20 | Qwen 2.5 Coder 32B Instructtext· SWE-rebench | 4% |
| 21 | GPT-4.1 nanotext· SWE-rebench | 0.5% |