Code Generation on LCB-IO
81.7Pass@1CWM RL
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CWM RLPublic pass=true2026.03 | 81.7 | — | 90.6 | — | |
| QWEN3-32BPublic pass=true2026.03 | 72.3 | — | 88.8 | — | |
| QWEN2.5-7B RLPublic pass=true2026.03 | 68 | — | 84.7 | — | |
| QWEN3-4BPublic pass=true2026.03 | 64.4 | — | 80.9 | — | |
| Execution RLEFfeedback=Oracle2026.03 | 63.8 | 70.9 | 72.8 | 88.5 | |
| Self-RLEF2026.03 | 62.3 | 70 | 71.9 | 87.1 | |
| CWM RL2026.03 | 61 | 67.6 | 69.2 | 82.9 | |
| CWMversion=Official2026.03 | 57.4 | 67.3 | 70.1 | 71.4 |