Kernel Generation on MUSA KernelBench Level 2
96Pass@8DeepSeek-V4-Pro
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DeepSeek-V4-ProModel Scale=V4-Pro2026.06 | 96 | 61.5 | 7.5 | |
| MusaCoder-27B-RLModel Scale=27B, Training Protocol=RL2026.06 | 96 | 84.2 | 10 | |
| GLM-5.1Model Scale=5.12026.06 | 93 | 68.1 | 8.9 | |
| MusaCoder-9B-RLModel Scale=9B, Training Protocol=RL2026.06 | 90 | 77 | 5.7 | |
| MusaCoder-27B-SFTModel Scale=27B, Training Protocol=SFT2026.06 | 84 | 64.5 | 4.9 | |
| MusaCoder-9B-SFTModel Scale=9B, Training Protocol=SFT2026.06 | 68 | 59 | 1.7 |