Code Generation on Reflective Verification Loop (test)
73.8All-Passed RateUnityMAS-O (3xQwen3-8B)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| UnityMAS-O (3xQwen3-8B)Setting=3xQwen3-8B, RL stage=Best after RL2026.05 | 73.8 | 0.448 | 154 | |
| UnityMAS-O (3xQwen3-4B)Setting=3xQwen3-4B, RL stage=Best after RL2026.05 | 68.6 | 0.431 | 169 | |
| 3xQwen3-8BSetting=3xQwen3-8B, RL stage=Before RL2026.05 | 29 | — | — | |
| 3xQwen3-4BSetting=3xQwen3-4B, RL stage=Before RL2026.05 | 25.5 | — | — |