GPU Kernel Code Generation and Optimization on KernelBench Level 2 (test)
19CorrectnessStitchCUDA-G
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| StitchCUDA-GHardware=PRO 6000 (BlackWell)2026.03 | 19 | 1.87 | 95 | |
| StitchCUDAHardware=PRO 6000 (BlackWell)2026.03 | 19 | 1.55 | 85 | |
| StitchCUDA-GHardware=H200 (Hooper)2026.03 | 19 | 1.79 | 90 | |
| CUDAForgeHardware=PRO 6000 (BlackWell)2026.03 | 18 | 1.28 | 80 | |
| StitchCUDA-QHardware=PRO 6000 (BlackWell)2026.03 | 18 | 0.77 | 45 | |
| CUDAForgeHardware=H200 (Hooper)2026.03 | 18 | 1.43 | 75 | |
| StitchCUDAHardware=H200 (Hooper)2026.03 | 18 | 1.82 | 85 | |
| StitchCUDA-KHardware=PRO 6000 (BlackWell)2026.03 | 17 | 0.82 | 65 | |
| StitchCUDA-QHardware=H200 (Hooper)2026.03 | 16 | 0.94 | 50 | |
| StitchCUDA-KHardware=H200 (Hooper)2026.03 | 15 | 0.77 | 60 | |
| Kevin32BHardware=PRO 6000 (BlackWell)2026.03 | 11 | 0.49 | 30 | |
| Claude-4-sonnetHardware=H200 (Hooper)2026.03 | 10 | 0.78 | 30 | |
| Kevin32BHardware=H200 (Hooper)2026.03 | 10 | 0.53 | 30 | |
| GPT5.2Hardware=H200 (Hooper)2026.03 | 9 | 0.63 | 30 | |
| GPT5.2Hardware=PRO 6000 (BlackWell)2026.03 | 8 | 0.66 | 30 | |
| Claude-4-sonnetHardware=PRO 6000 (BlackWell)2026.03 | 8 | 0.76 | 30 | |
| Qwen3-32BHardware=PRO 6000 (BlackWell)2026.03 | 0 | 0 | 0 | |
| Qwen3-32BHardware=H200 (Hooper)2026.03 | 0 | 0 | 0 |