KernelBench (kernelbench.com)
Coding agents write GPU kernels for real model workloads, such as fused MoE layers, paged attention and FP8 matrix multiplies. The site groups tasks into suites: Mega (one decode task), CUDA (4 tasks) and Hard (6 tasks). A run counts only if it passes an audit of correctness. Scores are speedup over a baseline or percent of the hardware roofline.
The site says it is not affiliated with Stanford's KernelBench. The almanac's Stanford record lists "KernelBench Hard" as an alias, and MiniMax M3's "KernelBench Hard" score may refer to this site's Hard suite instead, so that mapping needs a check. All linked runs are on an RTX PRO 6000. One CUDA result reads above 100% of roofline. About 99 GitHub stars.
- Made by
- Elliot Arledge (independent)
- Released
- July 2026
- Kind
- GPU kernels
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
No scores here yet
We have no scores for KernelBench (kernelbench.com) yet.
No lab model card reports it yet
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.
More in AI research
- PaperBench In use Agent replicates ICML 2024 papers from scratch. LLM judge grades against author-written hierarchical rubrics.
Qwen 3.8 Max 93% - MLS-Bench In use Agent must invent one algorithmic change in an ML research scaffold whose gain transfers across settings. Scored relative to human baselines. Claude Fable 5.1 50.3%
- PostTrainBench 1.2 In use CLI agent gets a base model, one H100 and a time budget to post-train it for a target benchmark. Scored on normalized gain. Claude Fable 5.1 44.6%
- MLE-bench In use Kaggle-style ML engineering competitions. Agent trains models and is scored against competition leaderboards. GPT-6 Astra 93.8%
- KernelBench In use Replace PyTorch operators with faster custom GPU kernels. Scored on correctness and speedup.
MiniMax-M3 28.8% - PMPP-Hard In use Agents write GPU kernels for 69 tasks that must be correct, and on 48 of them also run within 1.25 times the reference time. The score is the area under the solved-tasks curve over a 10-hour budget, out of 690 task-hours.
Inkling 2 Small 494.3