Mathematical Reasoning on AIME 24/25 (Yearly Scores)
59.3AIME 2024 ScoreDYPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DYPOBase Model=Qwen3-4B-Base2026.04 | 59.3 | 44 | |
| SFT -> RLBase Model=Qwen3-4B-Base2026.04 | 43.3 | 39.3 | |
| RLBase Model=Qwen3-4B-Base2026.04 | 40.6 | 37.3 | |
| SFTBase Model=Qwen3-4B-Base2026.04 | 33.3 | 27.3 | |
| Qwen3-4B-BaseBase Model=Qwen3-4B-Base2026.04 | 9.3 | 5.3 |