Mathematical Reasoning on AMC (pass@4)
87.31pass@4DAPO
Evaluation results
| Method | Links | |
|---|---|---|
| DAPOBackbone=Qwen3-4B, Training-free=false2026.04 | 87.31 | |
| ThinkTwiceBackbone=OLMo3-7B, Training-free=false2026.04 | 87.06 | |
| ThinkTwiceBackbone=Qwen3-4B, Training-free=false2026.04 | 85.54 | |
| DAPOBackbone=OLMo3-7B, Training-free=false2026.04 | 84.55 | |
| GRPOBackbone=OLMo3-7B, Training-free=false2026.04 | 84.48 | |
| Self-RefineBackbone=OLMo3-7B, Training-free=true2026.04 | 83.81 | |
| DrGRPOBackbone=Qwen3-4B, Training-free=false2026.04 | 82.66 | |
| Self-RefineBackbone=Qwen3-4B, Training-free=true2026.04 | 82.4 | |
| DrGRPOBackbone=OLMo3-7B, Training-free=false2026.04 | 82.32 | |
| GRPOBackbone=Qwen3-4B, Training-free=false2026.04 | 81.86 | |
| Base ModelBackbone=OLMo3-7B, Training-free=false2026.04 | 78.18 | |
| Base ModelBackbone=Qwen3-4B, Training-free=false2026.04 | 78.1 | |
| ReflexionBackbone=Qwen3-4B, Training-free=true2026.04 | 73.17 | |
| ReflexionBackbone=OLMo3-7B, Training-free=true2026.04 | 72.18 |