Kernel Generation on KernelBench Overall
25CR (Round 1)Codex
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CodexModel=GPT-5.22026.03 | 25 | 83 | 8 | 46 | |
| EvoKernelModel=Qwen3-Coder-30B2026.03 | 13 | 18 | 3 | 5.5 | |
| Pass@kModel=GPT-5.22026.03 | 13 | 24.5 | 5 | 11 | |
| Pass@kModel=Qwen3-Coder-30B2026.03 | 11 | 16 | 3.5 | 4 | |
| Pass@kModel=DeepSeek-V3.22026.03 | 11 | 23 | 3.5 | 4.5 | |
| EvoKernelModel=GPT-5.22026.03 | 11 | 98.5 | 4 | 83 | |
| RefinementModel=GPT-5.22026.03 | 10.5 | 71.5 | 4 | 22 | |
| RefinementModel=DeepSeek-V3.22026.03 | 9 | 35 | 0 | 6 | |
| RefinementModel=Qwen3-Coder-30B2026.03 | 6.5 | 11.5 | 1 | 3 | |
| EvoKernelModel=DeepSeek-V3.22026.03 | 5 | 29 | 1 | 9.5 |