evals.report
BenchmarksLabsCompareRun guides

KernelBench (Hard)

Stanford KernelBench's hardest tier: generate correct, high-performance GPU (CUDA) kernels from PyTorch reference operators, scored on the fraction of kernels that are both correct and faster than the baseline (fast₁).

Codingfast₁Higher is better
ModelLabScoreSource modelStatusDate
MiniMax M3MiniMax28.8%MiniMax M3VerifiedDetails

Each row reports the model’s fast₁ on KernelBench (Hard). Click a row for the full run context.