Mojo 1.1 opens its compiler to outside contributors for the first time

A programming language built to run AI faster than Python just became something anyone can improve. Mojo 1.1, the first update since the language achieved version 1.0 status, opens its compiler to contributions from outside the company that built it — a change that shifts Mojo from a corporate research project into a shared one.

Mojo was designed by Chris Lattner, who previously created the Swift programming language at Apple and spent years at Google working on AI infrastructure. The language looks like Python on the surface but runs tens of thousands of times faster for AI workloads, reaching hardware features like specialized processor registers that Python cannot easily access. It was built to address a specific frustration: Python is easy to write but slow on the numerical computation that drives modern AI models.

Under version 1.1, the Mojo compiler — the program that translates Mojo code into machine instructions — now accepts pull requests from external developers for the first time. The license has been Apache 2.0 since version 1.0, so the source code was already readable by anyone; what changed is that contributions can now flow back in, from the community into the language itself.

The release also introduces inferred member references, which reduce the amount of repetitive syntax programmers need to write when working with structured data. Compile times are shorter. The SIMD APIs — the low-level building blocks Mojo uses to reach the high-performance vector operations in modern processors — are now considered stable and suitable for production use.

Modular, the startup Lattner founded to build Mojo, was acquired by Qualcomm earlier this year, giving the language sustained corporate funding. The compiler opening comes alongside the simultaneous release of MAX 26.6, a framework for deploying AI models that runs on Mojo and added audio generation as a supported capability. Opening the compiler arrives after that institutional stability was in place, not before.

Whether the community will significantly shape Mojo’s direction depends on adoption, and adoption remains limited. Python’s ecosystem — its libraries, tutorials, and installed base of developers — is decades deep and unlikely to yield quickly to a faster alternative, even one with familiar syntax. Mojo can run Python code directly, which lowers the barrier to experimenting with it, but whether developers move production workflows onto Mojo is a question that remains largely open.

The first real test is what external contributors actually submit — whether those contributions reflect a community that has been running Mojo in production, or one still exploring at the edges. The changelog for Mojo 1.2 will be the first one anyone outside Qualcomm’s walls has genuinely helped write.

Tags: Open Source, AI development, Chris Lattner, Modular, Mojo, programming language