Code Generation on LiveCodeBench v6 (avg@4, pass@4)
47.14Avg@4Regular
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RegularBackbone=Qwen3.5-9B, Response Budget=32K-slot2026.05 | 47.14 | 54.2 | |
| PIPO-SFT + OPDBackbone=Qwen3.5-9B, Response Budget=32K-slot2026.05 | 46.56 | 62.6 | |
| RegularBackbone=Qwen3.5-4B, Response Budget=32K-slot2026.05 | 40.08 | 44.27 | |
| Eagle-2Backbone=Qwen3.5-9B, Response Budget=32K-slot2026.05 | 39.31 | 45.8 | |
| PIPO-SFTBackbone=Qwen3.5-9B, Response Budget=32K-slot2026.05 | 34.92 | 54.2 | |
| Eagle-2Backbone=Qwen3.5-4B, Response Budget=32K-slot2026.05 | 32.06 | 43.51 | |
| PIPO-SFT + OPDBackbone=Qwen3.5-4B, Response Budget=32K-slot2026.05 | 32.06 | 49.62 | |
| PIPO-SFTBackbone=Qwen3.5-4B, Response Budget=32K-slot2026.05 | 30.15 | 48.85 | |
| MTPBackbone=Qwen3.5-9B, Response Budget=32K-slot2026.05 | 18.32 | 31.3 | |
| MTPBackbone=Qwen3.5-4B, Response Budget=32K-slot2026.05 | 14.89 | 27.48 |