Mathematical Reasoning on AIME out-of-domain evaluation 2025
45.1Average Score @32GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GRPOModel=Qwen3-4B2026.06 | 45.1 | |
| GSPOModel=Qwen3-4B2026.06 | 42.08 | |
| DAPOModel=Qwen3-4B2026.06 | 39.06 | |
| WAPOModel=Qwen3-4B2026.06 | 38.43 | |
| DAPOModel=SmolLM3-3B2026.06 | 27.18 | |
| WAPOModel=SmolLM3-3B2026.06 | 26.77 | |
| GSPOModel=SmolLM3-3B2026.06 | 26.04 | |
| GRPOModel=SmolLM3-3B2026.06 | 25.93 | |
| GSPOModel=Gemma3-4B2026.06 | 14.27 | |
| DAPOModel=Gemma3-4B2026.06 | 13.75 | |
| GRPOModel=Gemma3-4B2026.06 | 13.54 | |
| WAPOModel=Gemma3-4B2026.06 | 12.71 |