Mojo Reached 1.0, Three Years After Promising to Make Python Fast
Modular declared its systems language stable on August 11. Nearly 200 outside contributors have shaped the standard library, and the compiler is still promised as open source before the year is out.
Mojo, the programming language built for AI workloads, officially hit version 1.0 on Tuesday, three years after its first public release. For a language, reaching 1.0 is less a technical event than a promise: from here on, the ground stops moving under your feet.
That promise is the whole point of this release. Modular, the company behind Mojo, had been changing the language quickly as it used it internally, and those changes made it painful for anyone outside the company to maintain a long-lived project. Version 1.0 draws a line. During the 1.x period, changes should mostly be additive. Breaking changes are still possible, but the company says they will be handled the way mature languages like C++ handle them, carefully and rarely.
The 26.5 release that carries Mojo 1.0 is mainly a tidying-up exercise. Where the language previously offered several ways to say the same thing, it now offers one: variables are always declared with var, closures have been unified, and there is a single pointer type instead of a family of them. Newer additions include Python-style lambda syntax, a much more stable language server so editors like VS Code behave properly, and memory safety diagnostics that catch a classic bug where adding to a list invalidates a reference into it.
What is behind it
Mojo exists because of an awkward split in AI work. Researchers write in Python because it is pleasant. The parts that actually need to be fast get written in C++ or CUDA by different people, and the two halves never quite fit together. Mojo’s pitch is that you can write both halves in one language that looks like Python but compiles down to code that runs well on CPUs, GPUs and accelerators. It is led by Chris Lattner, who previously created LLVM and Apple’s Swift language, which is a large part of why people took the project seriously from day one.
Two things are worth knowing before you get excited. Mojo is not open source yet. The standard library is, and nearly 200 contributors have landed over 1,100 pull requests touching more than 200,000 lines of it, but the compiler itself remains closed. Modular repeats its commitment to open sourcing the compiler and toolchain in 2026, which leaves a few months. And Modular was acquired by Qualcomm, so the language’s future now sits inside a chip company’s strategy.
What this means for you: If you do not write performance-critical code, this is background noise. If you do, a 1.0 is the signal that it is reasonable to start something you intend to keep. Install with uv pip install --upgrade mojo. If you are simply curious about where AI infrastructure is heading, watch whether that compiler actually opens this year. That is the commitment worth holding them to.
Sources
Source: https://www.modular.com/blog/modular-26-5-mojo-1-0-is-here
Nvidia's New Free Model Is Not the Smartest. It Is Just Very, Very Fast
Nemotron 3.5 Lightning matches OpenAI's gpt-oss-120b on intelligence scores with a quarter of the parameters, and produces around 670 tokens per second, the fastest in its comparison group.
