Kernel Code Generation on KernelBench Level-2
100AccuracyPS
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| PSPass Strategy=Multi-Pass, Steps=50, Skill Memory=true, Base Model=GPT-5-mini2026.04 | 100 | 2.12 | 63 | 23 | |
| IRPass Strategy=Multi-Pass, Steps=50, Skill Memory=false, Base Model=GPT-5-mini2026.04 | 100 | 1.96 | 44 | 20 | |
| IRPass Strategy=Multi-Pass, Steps=50, Skill Memory=true, Base Model=GPT-5-mini2026.04 | 100 | 2.59 | 67 | 33 | |
| OpenEvolvePass Strategy=Multi-Pass, Steps=50, Skill Memory=true, Base Model=GPT-5-mini2026.04 | 100 | 1.91 | 35 | 12 | |
| DR. KernelPass Strategy=Multi-Pass, Steps=4 x 14, Base Model=GPT-5-mini2026.04 | 100 | 1.78 | 51 | 13 | |
| AdaExplorePass Strategy=Multi-Pass, Steps=50, Base Model=GPT-5-mini2026.04 | 100 | 2.65 | 71 | 34 | |
| AdaExplorePass Strategy=Multi-Pass, Steps=100, Base Model=GPT-5-mini2026.04 | 100 | 3.12 | 78 | 44 | |
| AdaExplorePass Strategy=Multi-Pass, Steps=200, Base Model=GPT-5-mini2026.04 | 100 | 3.41 | 81 | 49 | |
| PSPass Strategy=Multi-Pass, Steps=50, Skill Memory=false, Base Model=GPT-5-mini2026.04 | 87 | 1.69 | 49 | 17 | |
| Claude-4.6-OpusPass Strategy=Single-Pass2026.04 | 60 | 0.76 | 16 | 5 | |
| GPT-5Pass Strategy=Single-Pass2026.04 | 51 | 0.78 | 19 | 4 | |
| DR. KernelPass Strategy=Single-Pass2026.04 | 46 | 0.86 | 14 | 6 | |
| AutoTritonPass Strategy=Single-Pass2026.04 | 41 | 0.44 | 7 | 1 | |
| GPT-5-miniPass Strategy=Single-Pass2026.04 | 22 | 0.34 | 9 | 3 |