Skip to main content
ModelScale

Coding benchmark

KernelBench leaderboard

KernelBench Hard H100. Every model the catalog carries a published KernelBench value for, ranked by that value.

CategoryCoding
MeasureMean peak fraction of hardware roofline over valid cells
Tasks6 GPU-kernel optimization problems
DifficultyAgentic GPU systems engineering
Published byKernelBench Hard

An agentic GPU-kernel benchmark that measures how much of the hardware roofline a model's correct, audit-clean kernels reach on six demanding CUDA and Triton problems.

No model in the catalog has a published KernelBench score.

The benchmark is defined by KernelBench Hard, but the catalog carries no value for it yet. An absent value is shown as absent here rather than as a zero.

All leaderboards · Coding capability leaderboard