Mathematical Reasoning on HMMT 02 26 (pass@1, pass@64)
11.9Pass@1CurveRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CurveRLBackbone=Qwen3-4B-Base2026.05 | 11.9 | 40.9 | |
| MaxRLBackbone=Qwen3-4B-Base2026.05 | 11.8 | 33.9 | |
| GRPOBackbone=Qwen3-4B-Base2026.05 | 9.9 | 27.2 | |
| GRPOBackbone=Qwen3-1.7B-Base2026.05 | 3.8 | 15.5 | |
| MaxRLBackbone=Qwen3-1.7B-Base2026.05 | 3.2 | 17.7 | |
| CurveRLBackbone=Qwen3-1.7B-Base2026.05 | 3 | 21 | |
| BaseBackbone=Qwen3-4B-Base2026.05 | 3 | 20.6 | |
| BaseBackbone=Qwen3-1.7B-Base2026.05 | 1.1 | 14.1 |