Code Generation on LiveCodeBench v6 (mean@8, pass@8)
61.82Mean@8MOPD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MOPDModel Backbone=Qwen3-8B2026.05 | 61.82 | 67.23 | |
| MOPDModel Backbone=Qwen3-4B2026.05 | 57.01 | 65.48 | |
| SDPOModel Backbone=Qwen3-8B2026.05 | 49.49 | 64.34 | |
| SDPOModel Backbone=Qwen3-4B2026.05 | 48.84 | 63.23 | |
| GRPOModel Backbone=Qwen3-8B2026.05 | 43.65 | 58.72 | |
| GRPOModel Backbone=Qwen3-4B2026.05 | 40.75 | 55.43 | |
| BaseModel Backbone=Qwen3-8B2026.05 | 30.97 | 53.02 | |
| BaseModel Backbone=Qwen3-4B2026.05 | 28.8 | 49.36 |