Mathematical Reasoning on MATH 500 (pass@1, pass@64)
85.1Pass@1MaxRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MaxRLBackbone=Qwen3-4B-Base2026.05 | 85.1 | 96.6 | |
| CurveRLBackbone=Qwen3-4B-Base2026.05 | 84.5 | 97.7 | |
| GRPOBackbone=Qwen3-4B-Base2026.05 | 82.8 | 94.7 | |
| MaxRLBackbone=Qwen3-1.7B-Base2026.05 | 72.1 | 94 | |
| CurveRLBackbone=Qwen3-1.7B-Base2026.05 | 71.6 | 94.6 | |
| GRPOBackbone=Qwen3-1.7B-Base2026.05 | 70.5 | 92.6 | |
| BaseBackbone=Qwen3-4B-Base2026.05 | 67.5 | 95.7 | |
| BaseBackbone=Qwen3-1.7B-Base2026.05 | 57.1 | 92.8 |