Kernel Generation on KernelBench
3.45SpeedupCuTeGen
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CuTeGenKernel=Softsign2026.04 | 3.45 | — | — | — | — | — | — | — | |
| CuTeGenKernel=Swish2026.04 | 2.45 | — | — | — | — | — | — | — | |
| CuTeGenKernel=ReLU2026.04 | 1.01 | — | — | — | — | — | — | — | |
| CuTeGenKernel=GELU2026.04 | 1.01 | — | — | — | — | — | — | — | |
| CuTeGenKernel=Softplus2026.04 | 1.01 | — | — | — | — | — | — | — | |
| CuTeGenKernel=Leaky ReLU2026.04 | 1 | — | — | — | — | — | — | — | |
| CuTeGenKernel=Sigmoid2026.04 | 1 | — | — | — | — | — | — | — | |
| CuTeGenKernel=Tanh2026.04 | 1 | — | — | — | — | — | — | — | |
| CuTeGenKernel=SELU2026.04 | 0.99 | — | — | — | — | — | — | — | |
| CuTeGenKernel=ELU2026.04 | 0.99 | — | — | — | — | — | — | — | |
| CuTeGenKernel=HardTanh2026.04 | 0.99 | — | — | — | — | — | — | — | |
| CuTeGenKernel=Hard Sigmoid2026.04 | 0.94 | — | — | — | — | — | — | — | |
| CuTeGenKernel=Softmax2026.04 | 0.88 | — | — | — | — | — | — | — | |
| CuTeGenKernel=LogSoftmax2026.04 | 0.87 | — | — | — | — | — | — | — | |
| EvoKernelPlatform=CUDA, # Ops=2502026.03 | — | 100 | 100 | 68 | — | — | — | — | |
| GPT 4o-mini2026.03 | — | 47 | — | 22 | 1.27 | 1.12 | 8 | 2.13 | |
| Kevin-32B2026.03 | — | 65 | — | 21 | 1.36 | 1.1 | 9 | 2.69 | |
| QwQ-32B2026.03 | — | 33 | — | 14 | 1.25 | 1.02 | 4 | 0.69 | |
| SEMREP2026.03 | — | 93 | — | 72 | 2.87 | 1.21 | 12 | 3.13 |