Mathematical Reasoning on AIME 26 (mean@8, pass@8)
70Mean@8DeepSeek-R1-Distill-Qwen-32B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DeepSeek-R1-Distill-Qwen-32B2026.05 | 70 | 83.33 | |
| SkyWork-OR1-Math-7B2026.05 | 62.08 | 80 | |
| GRPOStudent Model=DeepSeek-R1-Distill-Qwen-7B, Teacher Model=DeepSeek-R1-Distill-Qwen-32B2026.05 | 62.08 | 80 | |
| REOPOLDStudent Model=DeepSeek-R1-Distill-Qwen-7B, Teacher Model=DeepSeek-R1-Distill-Qwen-32B2026.05 | 54.17 | 80 | |
| LGRStudent Model=DeepSeek-R1-Distill-Qwen-7B, Teacher Model=DeepSeek-R1-Distill-Qwen-32B2026.05 | 53.75 | 80 | |
| LGR (topk)Student Model=DeepSeek-R1-Distill-Qwen-7B, Teacher Model=DeepSeek-R1-Distill-Qwen-32B2026.05 | 53.75 | 76.67 | |
| OPD (9k)Student Model=DeepSeek-R1-Distill-Qwen-7B, Teacher Model=DeepSeek-R1-Distill-Qwen-32B2026.05 | 52.5 | 76.67 | |
| OPD (topk)Student Model=DeepSeek-R1-Distill-Qwen-7B, Teacher Model=DeepSeek-R1-Distill-Qwen-32B2026.05 | 52.08 | 70 | |
| JSDStudent Model=DeepSeek-R1-Distill-Qwen-7B, Teacher Model=DeepSeek-R1-Distill-Qwen-32B2026.05 | 51.25 | 66.67 | |
| Base ModelStudent Model=DeepSeek-R1-Distill-Qwen-7B2026.05 | 42.92 | 63.33 | |
| LGRStudent Model=DeepSeek-R1-Distill-Qwen-1.5B, Teacher Model=SkyWork-OR1-Math-7B2026.05 | 34.17 | 63.33 | |
| OPD (topk)Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Teacher Model=SkyWork-OR1-Math-7B2026.05 | 33.75 | 66.67 | |
| LGR (topk)Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Teacher Model=SkyWork-OR1-Math-7B2026.05 | 33.75 | 63.33 | |
| JSDStudent Model=DeepSeek-R1-Distill-Qwen-1.5B, Teacher Model=SkyWork-OR1-Math-7B2026.05 | 32.5 | 50 | |
| OPD (9k)Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Teacher Model=SkyWork-OR1-Math-7B2026.05 | 32.08 | 56.67 | |
| REOPOLDStudent Model=DeepSeek-R1-Distill-Qwen-1.5B, Teacher Model=SkyWork-OR1-Math-7B2026.05 | 30 | 46.67 | |
| GRPOStudent Model=DeepSeek-R1-Distill-Qwen-1.5B, Teacher Model=SkyWork-OR1-Math-7B2026.05 | 28.75 | 50 | |
| DAPO w/ RRModel Backbone=Qwen3-8B-Base2026.05 | 27.5 | 43.33 | |
| DAPO w/ PR-RKLModel Backbone=Qwen3-8B-Base2026.05 | 25.83 | 40 | |
| DAPOModel Backbone=Qwen3-8B-Base2026.05 | 25.42 | 43.33 | |
| DAPO w/ PR-FKLModel Backbone=Qwen3-8B-Base2026.05 | 25 | 50 | |
| Base ModelStudent Model=DeepSeek-R1-Distill-Qwen-1.5B2026.05 | 17.92 | 46.67 | |
| DAPO w/ PR-FKLModel Backbone=Qwen3-4B-Base2026.05 | 17.08 | 30 | |
| DAPO w/ PR-RKLModel Backbone=Qwen3-4B-Base2026.05 | 15.83 | 33.33 | |
| DAPO w/ RRModel Backbone=Qwen3-4B-Base2026.05 | 15 | 30 | |
| DAPOModel Backbone=Qwen3-4B-Base2026.05 | 14.17 | 30 | |
| BaseModel Backbone=Qwen3-8B-Base2026.05 | 6.67 | 20 | |
| BaseModel Backbone=Qwen3-4B-Base2026.05 | 6.25 | 16.67 |