Mathematical Reasoning on OlympiadBench (Mean@8)
60.1Mean Score @8FlowRL
Evaluation Results
| Method | Links | |
|---|---|---|
| FlowRLBackbone=Qwen3-4B-Base2026.05 | 60.1 | |
| GSPOBackbone=Qwen3-4B-Base2026.05 | 57.2 | |
| DISABackbone=Qwen3-4B-Base2026.05 | 56.8 | |
| GRPOBackbone=Qwen3-4B-Base2026.05 | 47.5 | |
| FlowRLBackbone=Qwen2.5-7B2026.05 | 32.4 | |
| DISABackbone=Qwen2.5-7B2026.05 | 31.9 | |
| VanillaBackbone=Qwen3-4B-Base2026.05 | 30.1 | |
| GRPOBackbone=Qwen2.5-7B2026.05 | 29.1 | |
| GSPOBackbone=Qwen2.5-7B2026.05 | 28.8 | |
| VanillaBackbone=Qwen2.5-7B2026.05 | 23.9 |