Triton kernel generation on KernelBench LEVEL1 1.0
39.3Fast1DR. KERNEL-14B-STTS
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DR. KERNEL-14B-STTSscaling_method=sequential test-time scaling (STTS), selection_strategy=best turn selection2026.02 | 39.3 | 25.1 | 20.4 | 17.6 | |
| DR. KERNEL-14B-STTSscaling_method=sequential test-time scaling (STTS)2026.02 | 24.1 | 18.8 | 15.3 | 12.8 | |
| DR. KERNEL-14B2026.02 | 20.3 | 16.9 | 13.2 | 11.6 | |
| GPT-52026.02 | 19.5 | 16.5 | 12.5 | 11 | |
| GLM-4.72026.02 | 19.4 | 17.2 | 13.1 | 10.4 | |
| GPT-5Evaluation Engine=torch.compile2026.02 | 18.6 | 8 | 6.5 | 5.5 | |
| DR. KERNEL-14BEvaluation Engine=torch.compile2026.02 | 17.8 | 5 | 3.3 | 2.5 | |
| DR. KERNEL-8BEvaluation Engine=torch.compile2026.02 | 16 | 3 | 1.5 | 8.4 | |
| DR. KERNEL-8B2026.02 | 15.9 | 12.8 | 10.9 | 8.4 | |
| Claude-4.5-Sonnet2026.02 | 15.5 | 13.5 | 11 | 8.5 | |
| Claude-4.5-SonnetEvaluation Engine=torch.compile2026.02 | 10 | 2.2 | 2 | 1.8 | |
| Deepseek-V3.2-Thinking2026.02 | 7.5 | 5.5 | 4.5 | 4 | |
| Cold-Start-8Btraining=cold-start data2026.02 | 7.5 | 6.6 | 5 | 4.3 | |
| Qwen3-32B2026.02 | 6.1 | 4.9 | 4.3 | 4 | |
| Qwen3-Coder-A30BA32026.02 | 6 | 5.2 | 5.1 | 3.8 | |
| Qwen3-8B2026.02 | 5.8 | 4.8 | 4.1 | 3.4 | |
| AutoTriton2026.02 | 4.5 | 3.6 | 2.8 | 2.1 |