Mathematical Reasoning on Average (AIME24, AIME25, AMC23, MATH500, OlympiadBench)
61.03Pass@1MEL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MELBackbone=Qwen3-14B-Base2026.02 | 61.03 | 60.07 | 74.94 | |
| GRPOBackbone=Qwen3-14B-Base2026.02 | 56.3 | 56.47 | 73.24 | |
| MELBackbone=Qwen3-8B-Base2026.02 | 52.79 | 52.27 | 71.17 | |
| GRPOBackbone=Qwen3-8B-Base2026.02 | 48.43 | 51.07 | 67.33 | |
| MELBackbone=Qwen3-4B-Base2026.02 | 45.48 | 46.25 | 63.41 | |
| GRPOBackbone=Qwen3-4B-Base2026.02 | 41.56 | 44.92 | 61.04 | |
| BaselineBackbone=Qwen3-8B-Base2026.02 | 41.3 | 38.06 | 60.6 | |
| BaselineBackbone=Qwen3-14B-Base2026.02 | 41.21 | 37.26 | 62.34 | |
| BaselineBackbone=Qwen3-4B-Base2026.02 | 36.34 | 32.06 | 55.16 |