Kernel Generation on KernelBench Level 1
34Correctness Rate (Round 1)Codex
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CodexModel=GPT-5.22026.03 | 34 | 82 | 16 | 70 | |
| EvoKernelModel=Qwen3-Coder-30B2026.03 | 25 | 33 | 6 | 11 | |
| Pass@kModel=GPT-5.22026.03 | 24 | 36 | 9 | 19 | |
| Pass@kModel=Qwen3-Coder-30B2026.03 | 22 | 30 | 7 | 8 | |
| Pass@kModel=DeepSeek-V3.22026.03 | 21 | 33 | 7 | 9 | |
| EvoKernelModel=GPT-5.22026.03 | 20 | 97 | 7 | 90 | |
| RefinementModel=GPT-5.22026.03 | 19 | 88 | 7 | 41 | |
| RefinementModel=DeepSeek-V3.22026.03 | 16 | 44 | 0 | 12 | |
| RefinementModel=Qwen3-Coder-30B2026.03 | 13 | 22 | 2 | 6 | |
| EvoKernelModel=DeepSeek-V3.22026.03 | 9 | 39 | 2 | 19 |