Mathematical Reasoning on MATH (MH)
80.2Mean Harmonic Score (MH)GEAR
Evaluation results
| Method | Links | |
|---|---|---|
| GEARBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 80.2 | |
| OPSDBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 78.3 | |
| GEARBackbone=Qwen3-4B, Evaluation Context=Multi-Domain Generalization2026.05 | 77.6 | |
| ARPOBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 76.4 | |
| MT-GRPOBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 76.2 | |
| GRPOBackbone=Qwen3-4B, Evaluation Context=Multi-Domain Generalization2026.05 | 75.8 | |
| GRPOBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 75.4 | |
| OPSD+RLBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 74.6 | |
| ARPOBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 74.2 | |
| GEARBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 74 | |
| OPSDBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 72.8 | |
| GRPOBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 72.8 | |
| BaseBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 72.5 | |
| MT-GRPOBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 72.4 | |
| BaseBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 72.1 | |
| OPSD+RLBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 71.9 |