Mojo, the Python-like language for GPUs, is now fully open source

Modular has open-sourced the entire Mojo compiler and toolchain under the Apache 2.0 license with LLVM exceptions, finishing a plan it first promised in 2023. Mojo is a language with Python-like syntax built for high-performance GPU and systems programming, so it matters for machine learning: the aim is to let people write fast kernels without dropping down to C++ or CUDA. The standard library and hundreds of thousands of lines of kernel code were already public; this release adds the compiler itself, the piece the company had held back. The code lives in the modular/modular repository on GitHub and builds with Bazel.

Modular is candid about the limits. Community compiler contributions are not open yet, with that step targeted for the end of the year, and the core design team stays small on purpose to hold the language together. There is also a notable retreat: the team now says Mojo "may or may not evolve into a full superset of Python, and it's okay if it doesn't," a change from its original pitch as a Python superset. The company plans to lean on AI-assisted tooling to help developers port existing Python code instead.

Why it matters

If you write GPU or ML performance code, an Apache-licensed Mojo you can read, fork, and build on removes a major reason to hold off. The open compiler is the real unlock, but wait for outside contributions to be accepted before assuming the governance is truly open.