Mathematical Reasoning on AIME 25 (pass@1, maj@16, pass@16)
18.3Pass@1DiRL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DiRLBase Model=Qwen3-4B-Base2026.06 | 18.3 | 27 | 45.8 | |
| G2RLBase Model=Qwen3-4B-Base2026.06 | 15.6 | 25.7 | 42 | |
| EVOL-RLBase Model=Qwen3-4B-Base2026.06 | 12.3 | 24 | 35.1 | |
| Entropy BonusBase Model=Qwen3-4B-Base2026.06 | 12.1 | 23.5 | 37.6 | |
| GRPOBase Model=Qwen3-4B-Base2026.06 | 10.4 | 18.8 | 32.8 | |
| DiRLBase Model=Qwen3-1.7B-Base2026.06 | 7.3 | 11.7 | 28.3 | |
| G2RLBase Model=Qwen3-1.7B-Base2026.06 | 6 | 9.9 | 23.4 | |
| EVOL-RLBase Model=Qwen3-1.7B-Base2026.06 | 5.4 | 6.7 | 19.7 | |
| GRPOBase Model=Qwen3-1.7B-Base2026.06 | 4.4 | 7.8 | 21.6 | |
| Entropy BonusBase Model=Qwen3-1.7B-Base2026.06 | 4.2 | 7.4 | 22.7 |