Code Generation on LiveCodeBench (LCB) (Accuracy and Delta)
83AccuracyLLM-guided spec search
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LLM-guided spec searchStudent Model=Qwen3.5-9B, Optimization Status=Optimized, Teacher Selection=Best teacher per benchmark2026.05 | 83 | 34.5 | |
| LLM-guided spec searchStudent Model=Gemma4-E4B, Optimization Status=Optimized, Teacher Selection=Best teacher per benchmark2026.05 | 81.5 | 13.2 | |
| LLM-guided spec searchStudent Model=Qwen3.5-4B, Optimization Status=Optimized, Teacher Selection=Best teacher per benchmark2026.05 | 75 | 65 | |
| Gemma4-E4BStudent Model=Gemma4-E4B, Optimization Status=Base2026.05 | 68.3 | — | |
| LLM-guided spec searchStudent Model=Nemotron-Nano-4B, Optimization Status=Optimized, Teacher Selection=Best teacher per benchmark2026.05 | 67 | 57 | |
| Qwen3.5-9BStudent Model=Qwen3.5-9B, Optimization Status=Base2026.05 | 48.5 | — | |
| Nemotron-Nano-4BStudent Model=Nemotron-Nano-4B, Optimization Status=Base2026.05 | 10 | — | |
| Qwen3.5-4BStudent Model=Qwen3.5-4B, Optimization Status=Base2026.05 | 10 | — |