Mathematical Reasoning on AIME 2025 (Mean@8)
59.2Mean@8DISA
Evaluation Results
| Method | Links | |
|---|---|---|
| DISABackbone=Qwen3-4B-Base2026.05 | 59.2 | |
| FlowRLBackbone=Qwen3-4B-Base2026.05 | 53.3 | |
| GSPOBackbone=Qwen3-4B-Base2026.05 | 52.1 | |
| GRPOBackbone=Qwen3-4B-Base2026.05 | 40 | |
| DISABackbone=Qwen2.5-7B2026.05 | 13.3 | |
| FlowRLBackbone=Qwen2.5-7B2026.05 | 9.17 | |
| GRPOBackbone=Qwen2.5-7B2026.05 | 7.92 | |
| VanillaBackbone=Qwen3-4B-Base2026.05 | 6.25 | |
| GSPOBackbone=Qwen2.5-7B2026.05 | 4.17 | |
| VanillaBackbone=Qwen2.5-7B2026.05 | 3.33 |