Modular finally open-sources Mojo, three years after the promise, and it is no longer trying to be Python

Mojo is open source at last, and it is no longer trying to be Python

Modular kept its three-year-old promise on August 18, 2026, and the language that opened is not the one the promise was made about: the Mojo compiler and toolchain shipped under Apache 2.0, 39 months after the open-source pledge that came with the May 2023 launch 1 2. In between, around August 2025, Modular dropped Mojo's founding goal of becoming a full Python superset and repositioned the language as standalone and GPU-first, with syntax borrowed from Python 1. Promise kept, product changed.

What the repository actually discloses

Yesterday a developer outside Modular could run Mojo. Today they can read it, fork it, and build it: the compiler source sits in the repository under KGEN, and a single Bazel command downloads or builds everything needed for the compiler and the standard library 2 3. This is not a day-one skeleton. The repository carries more than 53,000 commits, and its compiler docs folder ships a walkthrough from a December 2025 internal onsite, an attic directory of older docs kept for "code archeology", and a README note that most of the docs assume you work in the Modular monorepo 4. The four closed years arrived as readable history.

Two things did not open in the same clone. The repository bundles Mojo under Apache 2.0 with LLVM exceptions together with MAX platform components, whose usage and distribution are licensed under the Modular Community License 3. The announcement's phrase for Mojo, "fully open source", is a claim about the language, not the platform around it 2. And the README accepts contributions to the standard library and MAX components today while declining compiler contributions, which the blog says Modular aims to change "by the end of this year" 3 2.

That split is what the release-day Hacker News argument was actually about. The top comment holds the release is, for now, source-available rather than open source, because upstreaming is closed, though the license already permits forks; others answer that Apache 2 plus complete source meets the open source definition 5. One commenter goes further, calling the no-upstream, SQLite-style model the future of open source, partly as a shield against "AI generated contributions" 5. The pairing is the tell: the same week Modular's migration story leans on AI assistants, its governance model is being defended as protection from them.

The superset died before the compiler opened

The pivot is the load-bearing fact, so here is the spine, dated:

• May 2023: Mojo launches promising open source and aiming to be a full Python superset, so existing Python code could bootstrap the ecosystem; launch-era coverage described a superset of Python from a team led by Chris Lattner, who previously created LLVM, Clang and Swift 1 6
• March 2024: the standard library opens, with commenters noting the compiler itself stays closed 7
• September 2024: an HN post argues the closed core, with no visible release date, is "preventing Mojo from crossing the chasm" 8
• February 2025: Modular's FAQ commits to open-sourcing Mojo in 2026, as recorded by a Hacker News submission linking the docs page 9
• August 2025: the superset goal goes: "Mojo may or may not evolve into a full superset of Python, and it's okay if it doesn't" 1
• June 24 and July 29, 2026: Qualcomm announces, then completes, its acquisition of Modular, per the post dates on Modular's blog index 2
• August 11, 2026: Mojo 1.0 ships with source stability 2
• August 18, 2026: the compiler and toolchain open under Apache 2.0 2

Run the dates and the sequencing is the story. The superset ambition was alive for roughly 27 of the 39 months between promise and delivery; the compiler opened about 12 months after that ambition was dropped, nearly 29 months after the standard library opened with the core still closed, and 20 days after Qualcomm's acquisition of Modular closed 1. The compiler did not open while the superset ambition was alive. Willison's summary is the honest release note: Mojo is now its own language, tuned to make GPU programming as painless as possible, with Python-inspired syntax that is not fully compatible with existing Python code 1.

The on-ramp is an AI, not a tutorial

A language moat has two parts: compatibility, so existing code keeps running, and sunk learning, so people stay with what they know. Dropping the superset gave away the first, so Mojo's migration bet now rests on machines absorbing the second. In August 2025 Modular wrote that it was encouraged by how well AI-assisted coding tools already handled Python-to-Mojo migration 1. That stance is now product: the FAQ's conversion answer leads with Mojo AI skills that help an AI coding assistant "translate Python code into working Mojo code", ahead of the alternative of building Python bindings 10. The porting capacity is not Modular's: skills steer assistants built by other companies, so Mojo competes to be the best port target rather than the only road in. If assistants make ports cheap, language adoption starts to get decided by which languages models port best, not by which languages have the best tutorials.