Mojo

The Modular Platform is an open and fully-integrated suite of AI libraries and tools that accelerates model serving and scales GenAI deployments. It abstracts away hardware complexity so you can run the most popular open models with industry-leading GPU and CPU performance without any code changes. To get started with the Modular Platform and serve a model using the MAX framework, see the quickstart guide. After your model endpoint is up and running, you can start sending the model inference requests using our OpenAI-compatible REST API. Explore all the models you can deploy with Modular in our Model Library. The MAX container is our Kubernetes-compatible Docker container for convenient deployment, which uses the MAX framework's built-in inference server. We have separate containers for NVIDIA and AMD GPU environments, and a unified container that works with both. For more information, see our MAX container docs or the Modular Docker Hub repository. To install Mojo and get started learning the language, see the Mojo quickstart. The Mojo site also features a comprehensive language guide with tutorials, a language reference, and API references. We're constantly open-sourcing more of the Modular Platform and you can find all of it in here. As of May, 2025, this repo includes over 450,000 lines of code from over 6000 contributors, providing developers with production-grade reference implementations and tools to extend the Modular Platform with new algorithms, operations, and hardware targets. It's quite likely the world's largest repository of open source CPU and GPU kernels ! Branches The main branch is in sync with the nightly build and subject to new bugs. Use this branch for contributions. MAX release branches are named max/vX.X . Mojo release branches are named mojo/vX.X.X (Mojo uses the PEP 440 version scheme). For stable releases prior to MAX 26.3 / Mojo 1.0.0b1, combined MAX/Mojo release branches are named modular/vX.X . We accept contributions to the Mojo standard library, MAX AI kernels, MAX model architectures, code examples, Mojo docs, and more. First, please read the Contribution Guide, and then refer to the following documentation about how to develop in the repo: /max/docs : Docs for developers working in the MAX framework codebase. /mojo/stdlib/docs : Docs for developers working in the Mojo standard library. We also welcome your bug reports. If you have a bug, please file an issue here. [2026/3] [Modular Platform 26.2][26.2] delivers state-of-the-art image generation with over 4x speedup on FLUX.2 models, expanded hardware support for NVIDIA B300, Jetson Thor, DGX Spark, and AMD RDNA consumer GPUs, and Mojo language upgrades that make it easier to write GPU kernels with AI coding agents. [2026/2] We announced that [BentoML is joining Modular][bentoml-joins]. We are committed to building in the open and will be extending our support of open source AI with [Bento's own open project][bentoml-github]. Read the answers in our [February 2026 AMA][bentoml-joins-ama] to learn more about our plans. [2026/1] [Modular Platform 26.1][26.1] graduates the MAX Python API out of experimental with PyTorch-like eager mode and model.compile() for production, stabilizes the MAX LLM Book, and expands Apple silicon GPU support. Mojo gains compile-time reflection, linear types, typed errors, and improved error messages as it progresses toward 1.0. [2025/12] We hosted our [Inside the MAX Framework Meetup][dec-meetup] reintroducing the MAX framework and taking the community through upcoming changes. [2025/11] [Modular Platform 25.7][25.7] provides a fully open MAX Python API, expanded hardware support for NVIDIA Grace superchips, improved Mojo GPU programming experience, and much more. [2025/11] We met with the community at [PyTorch 2025 + the LLVM Developers' Meeting][pytorch-llvm] to solicit community input into how the Modular platform can reduce fragmentation and provide a unified AI stack. [2025/09] [Modular Platfor