Mathematical Reasoning on HMMT 25 (pass@1, pass@64)
9Pass@1CurveRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CurveRLBackbone=Qwen3-4B-Base2026.05 | 9 | 32.4 | |
| MaxRLBackbone=Qwen3-4B-Base2026.05 | 7.9 | 27.8 | |
| GRPOBackbone=Qwen3-4B-Base2026.05 | 5.8 | 19 | |
| Zero-Shotn=N/A2026.05 | 4.16 | 28.06 | |
| DPólyan=82026.05 | 3.68 | 26.62 | |
| DPólyan=322026.05 | 3.61 | 25.55 | |
| DPólyan=162026.05 | 3.6 | 26.52 | |
| DSTaRn=N/A2026.05 | 2 | 28.84 | |
| CurveRLBackbone=Qwen3-1.7B-Base2026.05 | 1.7 | 19.3 | |
| DPólyan=42026.05 | 1.39 | 26.02 | |
| MaxRLBackbone=Qwen3-1.7B-Base2026.05 | 1.3 | 18.4 | |
| BaseBackbone=Qwen3-4B-Base2026.05 | 1 | 22.5 | |
| GRPOBackbone=Qwen3-1.7B-Base2026.05 | 0.5 | 10 | |
| BaseBackbone=Qwen3-1.7B-Base2026.05 | 0.2 | 10.6 |