KernelBench — leaderboard

250 PyTorch workloads (operators, fusions, full architectures) where LLMs generate CUDA kernels. Measures average speedup over PyTorch baseline on Nvidia L40S. From Stanford Scaling Intelligence Lab.

Metric: Avg Speedup vs PyTorch. Source: scalingintelligence.stanford.edu. Status: saturation imminent. 7 models tracked.

Top models

#ModelScore
1O11
2Claude 3.5 Sonnet0.87
3GPT-4o0.79
4Llama 3.1 405B0.7
5Gemini 1.5 Flash0.64
6Llama 3.1 70B0.54

Interactive version: theaggregate.ai/benchmark?slug=kernelbench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.