Mathematical Reasoning on AIME 2025 (pass@1, pass@64)
19.6Pass@1CurveRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CurveRLBackbone=Qwen3-4B-Base2026.05 | 19.6 | 55.6 | |
| GRPOBackbone=Qwen3-4B-Base2026.05 | 19.1 | 48.6 | |
| MaxRLBackbone=Qwen3-4B-Base2026.05 | 18.9 | 52.6 | |
| CurveRLBackbone=Qwen3-1.7B-Base2026.05 | 7.7 | 40.9 | |
| BaseBackbone=Qwen3-4B-Base2026.05 | 6.9 | 44.3 | |
| MaxRLBackbone=Qwen3-1.7B-Base2026.05 | 6.7 | 36.5 | |
| GRPOBackbone=Qwen3-1.7B-Base2026.05 | 4.8 | 29.3 | |
| BaseBackbone=Qwen3-1.7B-Base2026.05 | 3.1 | 34.6 |