Coding on LiveCodeBench hard v2
26.1Pass RateLocal Highest (Ours)
Evaluation Results
| Method | Links | |
|---|---|---|
| Local Highest (Ours)Selection Strategy=Local Highest2025.10 | 26.1 | |
| Global Highest (GRAPE)Selection Strategy=Global Highest2025.10 | 23.2 | |
| Original Qwen2.5-32B-InstructSelection Strategy=None (Base Model)2025.10 | 11.4 |