Benchmark
SWE-rebench 2025-04
Coding benchmarks compare models on programming tasks. Scores are percentages.
- Models measured
- 14
- Scores recorded
- 17
- Sources
- 1
- IQ weight
- 1.46
Score distribution
- 12.8%
- Mean
- 11.5%
- Median
- 6.9%
- Std dev
- 3.5%
- 10th percentile
- 24%
- 90th percentile
- 2.5%
- Min
This benchmark has a weight of 1.46 for chat models, measured over 14 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 | Gemini 2.5 Flash Preview 05-20 (no think)text, no-thinking· SWE-rebench | 25% |
| 2 | GPT-4.1 minitext· SWE-rebench | 24% |
| 3 | GPT 4.1text· SWE-rebench | 24% |
| 4 | DeepSeek V3text· SWE-rebench | 18.5% |
| 5 | DeepSeek V3tools· SWE-rebench | 17.5% |
| 6 | Qwen3 32Btext, no-thinking· SWE-rebench | 14.5% |
| 7 | Qwen 3 235B A22Btext, no-thinking· SWE-rebench | 14% |
| 8 | Qwen 3 235B A22Btext, thinking· SWE-rebench | 12.1% |
| 9 | Qwen3 32Btext, thinking· SWE-rebench | 11.5% |
| 10 | Gemini 2.0 Flashtext· SWE-rebench | 10.5% |
| 11 | Qwen2.5 72B Instructtext· SWE-rebench | 10% |
| 12 | Devstral Small 2505text, 2505· SWE-rebench | 9% |
| 13 | Llama 3.3 70B Instructtext· SWE-rebench | 9% |
| 14 | Llama 4 Mavericktext· SWE-rebench | 9% |
| 15 | Gemma 3 27B ITtext· SWE-rebench | 3.5% |
| 16 | Llama 4 Scouttext· SWE-rebench | 3% |
| 17 | Qwen 2.5 Coder 32B Instructtext· SWE-rebench | 2.5% |