Mathematical Reasoning on AMC (Pass@1 Accuracy, Length Exceeding Ratio)
85.2Pass@1 AccuracyGRPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRPOModel=Qwen3-4B-Instruct2026.01 | 85.2 | 0.7 | |
| -Model=Qwen3-4B-Instruct2026.01 | 84.5 | 33.9 | |
| GDPOModel=Qwen3-4B-Instruct2026.01 | 84.3 | 0.1 | |
| GDPOModel=DeepSeek-R1-7B2026.01 | 84 | 0.3 | |
| GRPOModel=DeepSeek-R1-7B2026.01 | 83.8 | 0.6 | |
| -Model=DeepSeek-R1-7B2026.01 | 82.9 | 57.2 | |
| GDPOModel=DeepSeek-R1-1.5B2026.01 | 69 | 2.3 | |
| GRPOModel=DeepSeek-R1-1.5B2026.01 | 64.5 | 3.2 | |
| -Model=DeepSeek-R1-1.5B2026.01 | 62 | 67.5 |