Mathematical Reasoning on AMC (pass@1, pass@16, maj@16)
61Pass@1DiRL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DiRLBase Model=Qwen3-4B-Base2026.06 | 61 | 75.6 | 94.5 | |
| G2RLBase Model=Qwen3-4B-Base2026.06 | 54.1 | 68.4 | 90.4 | |
| Entropy BonusBase Model=Qwen3-4B-Base2026.06 | 53.7 | 67.8 | 88.8 | |
| EVOL-RLBase Model=Qwen3-4B-Base2026.06 | 53.1 | 67.1 | 87.8 | |
| GRPOBase Model=Qwen3-4B-Base2026.06 | 53 | 66.3 | 86.4 | |
| DiRLBase Model=Qwen3-1.7B-Base2026.06 | 43.6 | 55.5 | 83.4 | |
| G2RLBase Model=Qwen3-1.7B-Base2026.06 | 40.5 | 53.8 | 82.5 | |
| Entropy BonusBase Model=Qwen3-1.7B-Base2026.06 | 38.4 | 52.9 | 74.4 | |
| EVOL-RLBase Model=Qwen3-1.7B-Base2026.06 | 38.2 | 53.1 | 79.4 | |
| GRPOBase Model=Qwen3-1.7B-Base2026.06 | 37.6 | 53 | 77.2 |