Output Prediction on LiveCodeBench (LCB) output-prediction
51Code Execution Success RateCPPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CPPOBase model=Qwen3.5-4B, K=42026.05 | 51 | 35 | 43 | |
| PKPOBase model=Qwen3.5-4B, K=42026.05 | 48 | 34 | 41 | |
| CPPOBase model=Qwen3.5-2B, K=42026.05 | 41 | 31 | 36 | |
| PKPOBase model=Qwen3.5-2B, K=42026.05 | 38 | 21 | 29.5 | |
| DirectBase model=Qwen3.5-4B, K=42026.05 | 38 | 29 | 33.5 | |
| Fixed-Strategy EnsembleBase model=Qwen3.5-4B, K=42026.05 | 32 | 30 | 31 | |
| DirectBase model=Qwen3.5-2B, K=42026.05 | 22 | 19 | 20.5 | |
| Fixed-Strategy EnsembleBase model=Qwen3.5-2B, K=42026.05 | 19 | 13 | 16 | |
| Plan-and-SolveBase model=Qwen3.5-2B, K=42026.05 | 16 | 10 | 13 | |
| Plan-and-SolveBase model=Qwen3.5-4B, K=42026.05 | 15 | 12 | 13.5 |