WHY MOJO

One language. Every accelerator.

Mojo gives you the feel of Python with the performance of a systems language. One language carries you from high level logic all the way down to the GPU kernel, compiled ahead of time with no interpreter in the way. You write real kernels without ever dropping into C++ or a dialect built for one vendor.

Python feel, systems speed

Write it like Python. It runs like hand tuned C++ or CUDA, with no interpreter overhead and no split between your prototype and your production code.

One language, top to bottom

Your host logic and your GPU kernel live in the same language and the same file. No jump into a separate CUDA or C++ toolchain just to go fast.

Portable by design

The same kernel is built to run on different GPUs and accelerators, so what you learn here is not tied to one vendor.

BORROWED FROM WHAT WORKS
  • Python syntax and the Python ecosystem, so it reads like code you already know.
  • The LLVM and MLIR compiler stack, the same foundation under Swift, Rust, and modern machine learning.
  • Ownership and value semantics in the spirit of Rust, for memory safety with no garbage collector.
  • First class SIMD and vectorization from the systems world.
NEW TO MOJO
  • The first language designed from the start for MLIR, which is what makes the hardware portability real.
  • One language that spans CPUs, GPUs, and AI accelerators instead of a bolt on kernel language.
  • Interop with the existing Python ecosystem, so a team can adopt it gradually.
  • Compile time metaprogramming with parameters, so kernels specialize to the hardware at build time.
BUILT BY THE PEOPLE WHO BUILT THE TOOLS
  • Modular was founded in 2022 by Chris Lattner and Tim Davis.
  • Chris Lattner created LLVM, Clang, and Swift, plus the MLIR compiler framework that Mojo is built on.
  • He led Developer Tools at Apple for over a decade, then worked on AI infrastructure at Google, Tesla, and SiFive.
  • At Google he led the TensorFlow infrastructure team and built XLA and MLIR, the compiler technology now under much of modern machine learning.
  • Tim Davis led machine learning products at Google, the APIs, compilers, and runtimes behind TensorFlow, as Lattner's product counterpart.

In short, the people designing Mojo already shipped the compiler foundations the rest of the industry runs on.

Start the mission  →See the curriculum