The reason to look at Mojo again in 2026 is that the central objection to it has been removed. For its first three years Mojo asked developers to invest in a language whose compiler was proprietary, with a promise of eventual opening that plenty of people reasonably declined to bet on. Modular did open it, in stages: the standard library in 2024, the kernels in 2025 and the compiler itself in 2026, under Apache 2.0 with LLVM exceptions, alongside a 1.0 release carrying source stability guarantees. That is the licence the LLVM and Swift communities use and it is the right one for a language.
Chris Lattner has now done this three times, with LLVM, with Swift and now with Mojo, and whatever you think of the language the governance instincts are sound. The manual reflects a language team rather than a marketing team. It opens with a quickstart that is a real tour of syntax and fundamentals, moves into the language chapters, and offers a Conway's Game of Life tutorial that walks you through building a complete command-line application. That choice matters more than it sounds.
Most language tutorials from AI infrastructure companies teach you to run a matrix multiplication faster than NumPy and stop, which teaches a benchmark rather than a language. Building a working program forces you through structs, traits, ownership, error handling and the standard library, which is where the actual learning is. The hard part is ownership and value semantics, and I would have liked the manual to dwell on it longer. Mojo has borrow checking concepts inherited from the Rust lineage but expressed differently, with argument conventions that determine whether a value is borrowed, mutably borrowed or transferred.
If you come from Python this is the genuinely unfamiliar material, and it is where every new Mojo programmer's first week of compiler errors comes from. The chapter is clear, but clear explanations of ownership rules are not what makes ownership click. Worked examples of getting it wrong are, and there are not enough of those. Now the part a review should be honest about.
Mojo is open source. MAX is not, in the same sense. The inference engine that provides most of Modular's commercial rationale is distributed under the Modular Community License, which is a source-available arrangement with usage conditions rather than an open source licence. The practical effect is that the language you are learning is free and the platform it was designed to feed is not, which is a familiar split and worth naming plainly rather than letting the word open cover both.
Qualcomm completed its acquisition of Modular in July 2026, an all-stock deal reported at roughly 3.9 billion dollars, with Lattner moving into an executive role running advanced AI software. Modular has said Mojo, MAX and Modular Cloud continue as products, and the Apache licence on the compiler means the language itself cannot be taken away regardless. The open question is direction rather than survival. A chip company acquiring a compiler company is buying software portability as a weapon against CUDA's moat, and the sensible expectation is that Qualcomm silicon becomes a first-class target faster than anyone else's.
That is not a bad outcome for a language whose whole pitch is heterogeneous hardware, and it does mean the neutrality argument needs watching. On the actual question of whether to spend a week here: it depends on why. If you want a language to get a job with, no. Mojo job listings barely exist and will not exist in volume for years, if ever.
If you want to understand why the Python-plus-C-plus-CUDA stack is a problem, and what a serious attempt to replace it with one language looks like, Mojo is the most interesting answer anyone has produced and the manual is a good way in. The ideas transfer even if the language does not win: compile-time metaprogramming over hardware targets, explicit ownership without a garbage collector, and a type system designed around accelerators rather than retrofitted to them. Four point two. A real language manual for a language that is finally open, taught through building something rather than benchmarking something, with an ecosystem too small to justify betting a career on and a new corporate owner whose incentives are worth watching.