SWE-rebench
Real GitHub issues collected every month after model release dates, so tasks are unlikely to be in training data. Nebius runs every model in the same tool-calling scaffold. Scores use the task window the site shows by default.
| Category | Coding |
|---|---|
| Metric | Resolved rate (%) |
| Source | SWE-rebench (original) |
| License | None stated |
| Benchmark page | swe-rebench.com |
| Status | Scored |
| Weight | 1.1 within Coding |
| Fit | Discrimination 0.66, noise 0.21 |
Results
| # | Model | Provider | Result | Date |
|---|---|---|---|---|
| 1 | Claude Fable 5 High | Anthropic | 64.5% | 2026-07-01 |
| 2 | Grok 4.5 High | xAI | 63.8% | 2026-07-01 |
| 3 | Claude Opus 5 High | Anthropic | 63.4% | 2026-07-01 |
| 4 | GLM-5.2 High | Zhipu AI | 62.9% | 2026-07-01 |
| 5 | GPT-5.6 Sol Medium | OpenAI | 62.3% | 2026-07-01 |
| 6 | Claude Sonnet 5 High | Anthropic | 56.8% | 2026-07-01 |
| 7 | MiniMax-M3 | MiniMax | 47.2% | 2026-07-01 |
| 8 | MiMo-V2.5-Pro | Xiaomi | 46.5% | 2026-07-01 |
| 9 | GPT-5.6 Luna Medium | OpenAI | 43.6% | 2026-07-01 |
| 10 | DeepSeek V4 Pro 0423 High | DeepSeek | 40.2% | 2026-07-01 |
| 11 | Qwen3.6 27B | Alibaba | 31.2% | 2026-07-01 |
| 12 | Qwen3.6 35B-A3B | Alibaba | 24.7% | 2026-07-01 |
| 13 | Qwen3.5 35B-A3B | Alibaba | 17.1% | 2026-07-01 |