Math Reasoning on BRUMO 2025 (pass@1, pass@64)
30.4Pass@1MaxRL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MaxRLModel Backbone=Qwen3-4B-Base2026.05 | 30.4 | 66.5 | |
| CurveRLModel Backbone=Qwen3-4B-Base2026.05 | 27.6 | 66.2 | |
| GRPOModel Backbone=Qwen3-4B-Base2026.05 | 27.3 | 47.1 | |
| BaseModel Backbone=Qwen3-4B-Base2026.05 | 18.2 | 46.7 | |
| CurveRLModel Backbone=Qwen3-1.7B-Base2026.05 | 18 | 45.3 | |
| MaxRLModel Backbone=Qwen3-1.7B-Base2026.05 | 15.9 | 42.5 | |
| GRPOModel Backbone=Qwen3-1.7B-Base2026.05 | 14.2 | 40.2 | |
| BaseModel Backbone=Qwen3-1.7B-Base2026.05 | 8 | 40.1 |