Today, we're excited to announce a major milestone for Modular: the complete MAX source code is now publicly available, alongside a radically simplified MAX License.

Developers can now explore, modify, and extend the entire MAX stack. Building on our open source kernels, models, and serving infrastructure, this release opens the remaining core components that power MAX's heterogeneous capabilities: the graph compiler, runtime, and hardware backends. These include support for CPUs and GPUs.

Under the new license, you can build and distribute commercial products with MAX without licensing fees or royalties. We’re also opening MAX to community contributions, starting with bug fixes and expanding over time. We’re also actively supercharging MAX with agentic infrastructure development that will increase our overall development approach.

This milestone builds on our progressive approach to opening our stack. We started with the Mojo standard library in 2024, followed by MAX's kernels, models, serving infrastructure, and APIs throughout 2025. In August of this year, we fully open sourced the entire Mojo compiler and toolchain. Today, we complete the picture by making the remaining MAX compiler, runtime, and hardware backends source available.

As Modular enters a new chapter within Qualcomm, we’re accelerating our mission to democratize heterogeneous compute. In the next few months, we will launch an alliance program, with a goal of uniting developers and hardware vendors around a single software platform. Opening MAX is the first step toward that shared ecosystem.

The new MAX License

We designed the new MAX License to balance two priorities: enabling the community to build with more flexibility, and balancing the long-term sustainability of MAX and its ecosystem.

That starts with giving developers the freedom to build. You can use, modify, extend, and build commercial products with MAX, without licensing fees or royalties. Our goal is to remove barriers to adoption and give developers, researchers, and hardware vendors the freedom to innovate on top of the entire stack.

Equally important is ensuring that the ecosystem can thrive over the long term. Our team has spent decades building and maintaining major open source projects, and we've seen the challenges they face. Fragmentation is a common pitfall, especially for foundational infrastructure like MAX. MAX’s core goal is to unify the programming model across hardware, and that goal becomes much harder to achieve if every platform develops its own incompatible version.

We designed the license with these considerations in mind. As with Mojo, we expect our approach to continue evolving as the technology matures and we hear more from the community.

What the MAX License covers

Many MAX components are already open source under Apache 2.0. These include the MAX kernel library in Mojo, Mojo interfaces for programming accelerators, Python APIs for the serving CLI and modeling, MAX models, and more. All of these components will remain under Apache 2.0.

The new MAX License covers the MAX internals we’re making source available today, plus the MAX binaries we distribute. Each source file includes a header identifying the license that applies to it.

Here’s what the MAX License allows:

• Unlimited MAX usage on any hardware, without licensing fees: Build with MAX on any hardware, for free.
• Contribute back to MAX: Fork MAX for the purposes of contributing back to the main MAX project.
• Build libraries and applications on top of MAX: Create projects and libraries using MAX and license them however you choose. Components of MAX, such as the kernels (Apache 2.0), the graph compiler (MAX License), and device backends (MAX License), retain their licenses and you can license your code as you see fit.

There a few limitations:

• AI-assisted reimplementation of MAX: You can't use any part of MAX as input to an AI system to produce a reimplementation of MAX. We want to build a community and pull people together, not be the foundation for many AI-driven forks.
• Commercial hosted inference services: If you use MAX for hosted inference or training services that you sell to customers, you must prominently display the “Modular Accelerated” wordmark on your product to recognize our contribution to your commercial products.
• Publication and distribution of modified source code: You can't publish or distribute modified MAX source code, except by forking MAX for the purpose of contributing changes back to the main project. We know that there are many fierce competitors in the AI landscape, and hope to replicate LLVM’s success of getting people to build on a shared codebase that can service many different companies' needs, better than any one can build for themself.

Our intent is to encourage experimentation and innovation while ensuring that MAX continues to evolve as a unified software foundation. Developers are free to modify and experiment with MAX, and we want to make it as easy as possible to contribute improvements back to the project. The license’s distribution restrictions are designed to prevent fragmentation into incompatible versions of MAX, not to discourage individual developers from exploring new ideas. We recognize that there are many legitimate reasons to experiment with modifications that aren’t ready for upstream contribution, and we’ll continue working with the community to make sure our development and contribution processes support that innovation.

MAX may also collect usage telemetry to help us improve the product. We handle personal information in accordance with our Privacy Policy, which describes users’ rights under applicable data protection laws, including the EU and UK GDPR and the California Consumer Privacy Act. As with most software licenses, MAX is provided as-is, without warranties.

For more details on the MAX License, explore the FAQ and read the full license.

What comes next

With the full MAX codebase now source available, we’re also opening MAX to community contributions starting today. Initially, we’ll accept bug fixes, with plans to expand to refactors and other changes over time, following the same gradual approach we took with the Mojo standard library and compiler.

Our next step is to bring MAX and Mojo to the broader AI ecosystem. At ModCon, we announced the Modular alliance program, bringing together hardware vendors, model labs, and data center operators around a common software foundation.

Partners who contribute engineering resources, compute, or early model checkpoints will help shape the future of MAX, with a voice in governance and conformance standards, input into the roadmap, priority for upstream contributions, early access to model checkpoints, ownership of dedicated hardware backends, and more permissive licensing.

We’ll share more about the alliance in the coming months. If you’re interested in getting involved, reach out at alliance@modular.com.

To learn more about the new MAX License, join Modular cofounders Chris Lattner and Tim Davis for a live Q&A on November 10 at 10 AM PT. RSVP here.

Get started

• Browse the MAX source on GitHub
• Read the full MAX License
• Read the MAX License FAQ
• Install MAX with pip install max[all]
• Check out the contributor guide to submit a bug fix
• Build MAX from source
• Share your feedback on the Modular forum or in Discord

We look forward to seeing what you build with MAX. If any part of the license remains unclear, we want to hear from you. Your feedback shaped this release, and it will continue to guide the steps that follow.