Mathematical Reasoning on Average (AIME 2025, BeyondAIME, HMMT 02/25, HMMT 02/26, MATH-500)
26.7Pass@1CurveRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CurveRLBackbone=Qwen3-4B-Base2026.05 | 26.7 | 53.4 | |
| MaxRLBackbone=Qwen3-4B-Base2026.05 | 26.4 | 48.7 | |
| GRPOBackbone=Qwen3-4B-Base2026.05 | 25.1 | 43.7 | |
| CurveRLBackbone=Qwen3-1.7B-Base2026.05 | 17.1 | 39.9 | |
| MaxRLBackbone=Qwen3-1.7B-Base2026.05 | 16.9 | 37.7 | |
| BaseBackbone=Qwen3-4B-Base2026.05 | 16.5 | 41.4 | |
| GRPOBackbone=Qwen3-1.7B-Base2026.05 | 16.1 | 32.6 | |
| BaseBackbone=Qwen3-1.7B-Base2026.05 | 12.5 | 33.8 |