Code Generation on LiveCodeBench v1-5
30.6pass@1SAGE
Evaluation Results
| Method | Links | |
|---|---|---|
| SAGEBackbone=Qwen-3-4B-Base2026.03 | 30.6 | |
| SAGEBackbone=Qwen-2.5-7B-Instruct2026.03 | 26.4 | |
| AZRBackbone=Qwen-3-4B-Base2026.03 | 26.1 | |
| AZRBackbone=Qwen-2.5-7B-Instruct2026.03 | 25.3 | |
| MAEBackbone=Qwen-3-4B-Base2026.03 | 24.2 | |
| MAEBackbone=Qwen-2.5-7B-Instruct2026.03 | 23.3 | |
| Base ModelBackbone=Qwen-3-4B-Base2026.03 | 21.5 | |
| Base ModelBackbone=Qwen-2.5-7B-Instruct2026.03 | 17.5 | |
| SAGEBackbone=Qwen-2.5-3B-Instruct2026.03 | 16.9 | |
| MAEBackbone=Qwen-2.5-3B-Instruct2026.03 | 15.9 | |
| AZRBackbone=Qwen-2.5-3B-Instruct2026.03 | 15 | |
| Base ModelBackbone=Qwen-2.5-3B-Instruct2026.03 | 12 |