Mathematical Reasoning on AMC 2023 (Avg@32, Pass@32)
96.3Avg@32HTPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HTPOBackbone=Qwen3-8B-Instruct2026.05 | 96.3 | — | |
| GSPOBackbone=Qwen3-8B-Instruct2026.05 | 94.1 | — | |
| SAPOBackbone=Qwen3-8B-Instruct2026.05 | 94 | — | |
| DAPOBackbone=Qwen3-8B-Instruct2026.05 | 93.8 | — | |
| 80/20-RuleBackbone=Qwen3-8B-Instruct2026.05 | 93.8 | — | |
| BAPOBackbone=Qwen3-8B-Instruct2026.05 | 93.3 | — | |
| GRPO†Backbone=Qwen3-8B-Instruct2026.05 | 92.2 | — | |
| HTPOBackbone=Qwen3-8B-Base2026.05 | 83 | — | |
| GSPOBackbone=Qwen3-8B-Base2026.05 | 82 | — | |
| SAPOBackbone=Qwen3-8B-Base2026.05 | 81.9 | — | |
| SCOPE (Ours)Teacher Model=SkyWork-OR1-Math-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B2026.04 | 80.9 | 97.2 | |
| BAPOBackbone=Qwen3-8B-Base2026.05 | 77.8 | — | |
| 80/20-RuleBackbone=Qwen3-8B-Base2026.05 | 77.2 | — | |
| w/ OPDTeacher Model=SkyWork-OR1-Math-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B2026.04 | 75.9 | 95 | |
| DAPOBackbone=Qwen3-8B-Base2026.05 | 75.9 | — | |
| w/ GRPOTeacher Model=SkyWork-OR1-Math-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B2026.04 | 75.1 | 95 | |
| R1-Distill-Qwen-1.5BTeacher Model=SkyWork-OR1-Math-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B2026.04 | 72.7 | 94.7 | |
| GRPO†Backbone=Qwen3-8B-Base2026.05 | 70.2 | — | |
| w/ KDTeacher Model=SkyWork-OR1-Math-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B2026.04 | 69.1 | 96.3 | |
| SCOPE (Ours)Teacher Model=Qwen3-8B-Instruct, Student Model=Qwen3-1.7B-Base2026.04 | 46.3 | 83 | |
| w/ OPDTeacher Model=Qwen3-8B-Instruct, Student Model=Qwen3-1.7B-Base2026.04 | 43.1 | 80.1 | |
| w/ GRPOTeacher Model=Qwen3-8B-Instruct, Student Model=Qwen3-1.7B-Base2026.04 | 33.1 | 60 | |
| w/ KDTeacher Model=Qwen3-8B-Instruct, Student Model=Qwen3-1.7B-Base2026.04 | 29.8 | 66.1 | |
| Qwen3-1.7B-BaseTeacher Model=Qwen3-8B-Instruct, Student Model=Qwen3-1.7B-Base2026.04 | 24.6 | 69.4 |