NVIDIA ComputeEval - cuda-kernels: leaderboard
Metric: Pass@1 (%, zero-shot, 2026.1 release, 114 problems). Source: huggingface.co. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (High) | 87.7 |
| 2 | GPT-5.4 (Medium) | 85.1 |
| 3 | GPT-5 Mini (High) | 76.3 |
| 4 | GLM-5 (Thinking) | 73.7 |
| 5 | Claude Opus 4.6 (High) | 72.8 |
| 6 | Kimi K2.5 (Thinking) | 71.9 |
| 7 | Gemini 3.1 Flash Lite (High) | 71.9 |
| 8 | Gemini 3.1 Flash Lite (Medium) | 71.9 |
| 9 | GPT-OSS-120B (High) | 71.1 |
| 10 | Claude Sonnet 4.6 (High) | 70.2 |
| 11 | Claude Sonnet 4.6 (Medium) | 70.2 |
| 12 | Claude Opus 4.6 (Medium) | 70.2 |
| 13 | GPT-5 Mini (Medium) | 68.4 |
| 14 | GPT-OSS-120B (Medium) | 67.5 |
| 15 | GPT-OSS-20B (High) | 61.4 |
Interactive version: theaggregate.ai/benchmark?slug=nvidia-computeeval-cuda-kernels · How It Works · Data refreshed daily, snapshot 2026-09-19.