Mojo 1.0 is stable — the compiler is still closed

Mojo 1.0 shipped today. The language Chris Lattner started to give AI code Python's syntax and Rust's memory model finally has a version number that means something: Modular is promising API stability across the 1.x line. That is the part that lets you plan. The part that should keep you from betting a product on it is that the compiler you would be compiling with is still closed source, and it now belongs to Qualcomm.

What actually happened

Modular's 26.5 release marks Mojo 1.0. The stability language is careful and worth reading literally: during the 1.x timeframe changes "should primarily be additive," and breaking changes "may still be made, but will be managed with care, following the standards of how mature languages (e.g. C++) evolve over time." That is a real commitment, not a guarantee of zero churn. The release itself is mostly cleanup — Python-style lambda syntax for inline closures, a single unified Pointer type, memory-safety diagnostics for reference invalidation, and a more stable LSP server for VS Code. You upgrade with uv pip install --upgrade mojo.

The standard library has been open source with outside contributors for a while. The compiler and toolchain have not. Modular restates the commitment in the same post: it will open source the Mojo compiler and toolchain in 2026. The Register notes the language site teases that this could happen at ModCon next week — that's a tease, not a date, so treat it as unscheduled.

The ownership changed underneath all of this. Qualcomm acquired Modular in June in an all-stock deal worth roughly $3.9 billion. A vendor-neutral "run it on any chip" toolchain is now developed inside a chip company, and the compiler that decides what "any chip" means in practice is the piece that hasn't been published yet.

Why it matters for your business

If you write ordinary business software, Mojo 1.0 changes nothing this quarter. Its niche is the layer under your AI calls — kernels, inference servers, the code that has to be fast on specific silicon. Most of what we build for clients calls an inference endpoint and never touches that layer.

Where it does matter is the second-order effect. Mojo exists to break CUDA lock-in, and the price of your tokens is partly set by how much competition Nvidia has at the software layer. A genuinely open Mojo compiler pressures that. A stable-but-closed one owned by a chipmaker mostly pressures it in Qualcomm's direction.

So the rule we apply to any language, runtime, or framework holds here: adopt on the license you can read, not the roadmap you've been told. If you're evaluating Mojo for a real workload, the trigger is the compiler open-sourcing under a license that survives an owner change — not the 1.0 tag. Until then, keep it at the edge of your system where you can rip it out: a compiled component behind a stable interface, not the language your business logic lives in.

Sources: Modular, The Register.