KernelBench: Can LLMs Write Efficient GPU Kernels?
Frontier
A benchmark for evaluating the ability of LLMs to write efficient GPU kernels.
- Publisher
- Stanford University
- Domain
- Coding
- License
- mit
- Published
- Feb 2026
- Notable for
- Benchmark for evaluating Coding.
Cite
Notes
Only stored in your browser.
Attribution
- README
- github.com/UKGovernmentBEIS/inspect_evals/blob/main/src/inspect_evals/kernelbench/eval.yamlMIT
- Leaderboard scores
- prime-hub
Top score 0.0% by DeepSeek V3.1 - 3 models reporting (3 frontier)
Top models
3
Related tools
5Implementations, trainers, datasets and scaffolds linked to this eval.
FAQ
- What is KernelBench: Can LLMs Write Efficient GPU Kernels??
- A benchmark for evaluating the ability of LLMs to write efficient GPU kernels.
- What is the current top score on KernelBench: Can LLMs Write Efficient GPU Kernels??
- The top reported score is 0.0% by DeepSeek V3.1, across 3 models reporting (3 from frontier labs).
- How can a model improve its KernelBench: Can LLMs Write Efficient GPU Kernels? score?
- Tools linked to KernelBench: Can LLMs Write Efficient GPU Kernels? on Sophon include Kernelbench RL Env (Community), Kernelbench RL Env (Community), Kernelbench RL Env (Prime Intellect), Kernelbench Kguard RL Env (Community) - RL environments, datasets, and scaffolds that target this eval.
- What license is KernelBench: Can LLMs Write Efficient GPU Kernels? under?
- KernelBench: Can LLMs Write Efficient GPU Kernels? is available under mit.