Modular is now accepting outside contributions to the Mojo compiler. The MAX 26.6 release notes put it in one line: "we're now accepting contributions to the Mojo compiler!" That closes the gap this site reported when the compiler went open source under Apache 2.0 with contributions still shut in August.
At that point Modular said it aimed to start taking compiler contributions by late 2026. It arrived early, and it arrived with a release that also deletes syntax.
What actually opened
The contribution notice sits on Modular's GitHub release page rather than in the language's own release notes, which is worth knowing if you go looking for the process.
Neither page names a contributor licence agreement or a review policy. So the invitation is public while the mechanics are not yet written down, and anyone planning to send a compiler patch should expect to ask.
The removals in 1.1
This release finishes a migration rather than starting one. The Mojo 1.1.0 notes say the release "completes the removal of APIs deprecated during the v1.0 cycle", and list what went:
| Removed | Replacement |
|---|---|
fn keyword |
Replaced during the 1.0 cycle |
alias keyword |
Replaced during the 1.0 cycle |
__comptime_assert |
Replaced during the 1.0 cycle |
@parameter if |
comptime if |
@parameter for |
comptime for |
The notes also record that the @parameter decorator on parametric closures is renamed @__parameter, and is now used only for declaring legacy closures.
The important word is "completes". These constructs were replaced in 1.0, so 1.1 is where the old spellings stop working. If you skipped that migration because the old forms still compiled, this is the release that ends the grace period.
Some foundations also settled. The notes say Modular has "marked more APIs in the standard library as stable, such as String, SIMD, and List", naming specific signatures on each.
What shipped alongside it
Mojo 1.1 rides the same version train as MAX 26.6, Modular's inference platform, and the platform is where the performance numbers are. The release notes claim the kernel speedups in the chart above, including 4.8 times faster Gemma 4 decode attention and 16 times faster low-batch NVFP4 quantization on Nvidia B200.
Two structural changes matter more than the multipliers. The max.gpu package now covers everything that used to live in std.gpu, and MAX gained audio generation, producing "44.1 kHz stereo music from a text prompt" using MiniMax-Music3. Phoronix reports the kernel work spans both AMD and Nvidia GPUs.
What this means for developers
Pin your Mojo version before you upgrade anything. A release that removes fn, alias and two @parameter forms will not compile code written against the pre-1.0 spellings. The error will come from your source rather than your dependencies. Search your tree for those five constructs first, then upgrade.
If you write GPU kernels against Modular's stack, the std.gpu to max.gpu move is the import change to schedule. That is a mechanical rename, but it is the kind that touches every file in a kernel library.
The stability markers are the useful planning signal. Modular marking String, SIMD and List stable says those signatures are safe to build on. That is a different promise from the language sitting at 1.x. Treat anything not on that list as still moving.
On contributions, temper the expectation. Qualcomm's purchase of Modular closed in July 2026, the compiler opened in August, and outside patches became welcome in September. That is a fast sequence, and the absence of a published CLA or review process suggests the door opened before the paperwork was finished. Useful if you want to fix a bug you have been carrying, less so if you need to know in advance how a large change gets accepted.
This article was first published on Tech AI Wire.
Also available in
Deutsch · 日本語 · Français · Español · Português
Top comments (0)