Code Generation on LiveCodeBench v6 (test) (Pass@4)
74.8Pass@4CPPO
Evaluation Results
| Method | Links | |
|---|---|---|
| CPPOBase model=Qwen3.5-9B2026.05 | 74.8 | |
| PKPOBase model=Qwen3.5-9B2026.05 | 58.8 | |
| PlanSearchBase model=Qwen3.5-9B2026.05 | 56.4 | |
| UpSkillBase model=Qwen3.5-9B2026.05 | 54.2 | |
| CPPOBase model=Qwen3.5-4B2026.05 | 53.5 | |
| Tuple Planner SFTBase model=Qwen3.5-9B2026.05 | 53.4 | |
| Pass@K Training / RLVRBase model=Qwen3.5-9B2026.05 | 50.3 | |
| Plan-and-SolveBase model=Qwen3.5-9B2026.05 | 49 | |
| PKPOBase model=Qwen3.5-4B2026.05 | 48.8 | |
| Direct SolveBase model=Qwen3.5-9B2026.05 | 48.1 | |
| UpSkillBase model=Qwen3.5-4B2026.05 | 41.1 | |
| Tuple Planner SFTBase model=Qwen3.5-4B2026.05 | 36.8 | |
| Pass@K Training / RLVRBase model=Qwen3.5-4B2026.05 | 31.6 | |
| Plan-and-SolveBase model=Qwen3.5-4B2026.05 | 25.5 | |
| PlanSearchBase model=Qwen3.5-4B2026.05 | 23.6 | |
| Direct SolveBase model=Qwen3.5-4B2026.05 | 21.4 |