CUTLASS

license

BSD-3-Clause(OSI)

source

public

core features withheld

no

The C++ library is public under a BSD-3-Clause LICENSE file with no paid or gated edition. The CuTe DSL added in CUTLASS 4 is a separate case, since its files carry an NVIDIA proprietary license.

• https://raw.githubusercontent.com/NVIDIA/cutlass/main/CMakeLists.txt recorded 2026-09-30

  Top-level CMakeLists.txt defines the library from the repository as "add_library(CUTLASS INTERFACE)", a header-only target built with plain options and no key or edition switch. Searched the body for enterprise, license key, paid and proprietary: none.

• https://raw.githubusercontent.com/NVIDIA/cutlass/main/LICENSE.txt recorded 2026-09-30

  LICENSE file opens with an NVIDIA copyright line, declares "SPDX-License-Identifier: BSD-3-Clause" and then runs the BSD text beginning "Redistribution and use in source and binary forms, with or without".

• https://raw.githubusercontent.com/NVIDIA/cutlass/main/README.md recorded 2026-09-30

  README About section states "CUTLASS is released by NVIDIA Corporation as Open Source software under the" 3-clause BSD license, and the library is described as "a collection of C++ CUDA template abstractions". Searched the body for enterprise, paid, license key and proprietary: none.

PyPI downloads of `nvidia-cutlass` are the adoption signal here, though CUTLASS is a header-only template library used from source in most cases, so this figure is a floor on real use rather than a full measure of it. It is still the project's own distribution channel.

• https://pypistats.org/api/packages/nvidia-cutlass/recent recorded 2026-10-07

  135,749 downloads in the trailing 30 days for nvidia-cutlass (pypistats last_month)

PyTorch and FlashAttention each vendor CUTLASS directly as a submodule and build against its headers. It covers one layer of the kernel stack rather than the full path from a framework graph to a target, which keeps it one step below Apache TVM.

• https://raw.githubusercontent.com/Dao-AILab/flash-attention/main/.gitmodules recorded 2026-09-30

  FlashAttention .gitmodules vendors CUTLASS with "path = csrc/cutlass" and "url = https://github.com/NVIDIA/cutlass.git".

• https://raw.githubusercontent.com/NVIDIA/cutlass/main/README.md recorded 2026-09-30

  README describes CuTe as "a collection of C++ CUDA template abstractions for" tensors of threads and data, states "CUTLASS 3.0 and beyond adopts CuTe throughout the GEMM hierarchy in its templates." and "This release has been tested against the following packages:" listing FlashAttention, Pytorch and TensorRT-LLM.

• https://raw.githubusercontent.com/pytorch/pytorch/main/.gitmodules recorded 2026-09-30

  PyTorch .gitmodules vendors CUTLASS with "path = third_party/cutlass" and "url = https://github.com/NVIDIA/cutlass.git".