Math Reasoning on Minerva Math (pass@1, pass@64)
42.8pass@1MaxRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MaxRLModel Backbone=Qwen3-4B-Base2026.05 | 42.8 | 62.3 | |
| GRPOModel Backbone=Qwen3-4B-Base2026.05 | 40.8 | 58.3 | |
| CurveRLModel Backbone=Qwen3-4B-Base2026.05 | 39.9 | 62.9 | |
| GRPOModel Backbone=Qwen3-1.7B-Base2026.05 | 31.4 | 57.7 | |
| MaxRLModel Backbone=Qwen3-1.7B-Base2026.05 | 31.2 | 59.5 | |
| CurveRLModel Backbone=Qwen3-1.7B-Base2026.05 | 30.3 | 60.3 | |
| BaseModel Backbone=Qwen3-4B-Base2026.05 | 25.9 | 61.3 | |
| BaseModel Backbone=Qwen3-1.7B-Base2026.05 | 20.8 | 56.9 |