SWE-rebench Leaderboard (default window 15/05/2026–01/07/2026, 111 problems)
SWE-rebench Leaderboard (default window 15/05/2026–01/07/2026, 111 problems) is run by SWE-rebench (Nebius). It has ranked 17 models, of which the catalogue holds 17, scoring from 17.1 to 64.5 on Resolved Rate (%).
measured by SWE-rebench (Nebius) · the board itself
| Place | Model | Metric | Score |
|---|---|---|---|
| 1stof 17 | Claude Fable 5 | Resolved Rate (%) (high) | 64.5 |
| 2ndof 17 | Grok 4.5 | Resolved Rate (%) (high) | 63.8 |
| 3rdof 17 | Claude Opus 5 | Resolved Rate (%) (high) | 63.4 |
| 4thof 17 | GLM 5.2 | Resolved Rate (%) (high) | 62.9 |
| 5thof 17 | GPT-5.6 Sol | Resolved Rate (%) (medium) | 62.3 |
| 6thof 17 | Junie | Resolved Rate (%) | 61.8 |
| 7thof 17 | Claude Code | Resolved Rate (%) | 60.4 |
| 8thof 17 | Codex | Resolved Rate (%) | 58 |
| 9thof 17 | Claude Sonnet 5 | Resolved Rate (%) (high) | 56.8 |
| 10thof 17 | Cursor Agent | Resolved Rate (%) | 51.7 |
| 11thof 17 | MiniMax M3 | Resolved Rate (%) | 47.2 |
| 12thof 17 | MiMo-V2.5-Pro | Resolved Rate (%) | 46.5 |
| 13thof 17 | GPT-5.6 Luna | Resolved Rate (%) (medium) | 43.6 |
| 14thof 17 | DeepSeek V4 Pro | Resolved Rate (%) (high) | 40.2 |
| 15thof 17 | Qwen3.6 27B | Resolved Rate (%) | 31.2 |
| 16thof 17 | Qwen3.6 35B A3B | Resolved Rate (%) | 24.7 |
| 17thof 17 | Qwen3.5-35B-A3B | Resolved Rate (%) | 17.1 |
Read from the board on 2026-09-16