Mojo goes fully open source — compiler, toolchain, Apache 2.0

Chris Lattner's GPU-first language ships its whole stack on GitHub, right after Qualcomm bought it. Plus: Qwen's 27B ties GPT-5.6 Luna, Copilot cuts 6 models.

Nowline AUG 19 8:00 AM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • The compiler is in the box this time

    Modular open-sourced the entire Mojo stack — compiler, toolchain, and standard library — under Apache 2.0, not just the stdlib it opened back in 2024. Clone github.com/modular/modular and a single Bazel command builds and runs it: `./bazelw run --config=build-mojo KGEN:mojo -- run hello.mojo`.

  • Built by the person behind LLVM and Swift

    Mojo is Chris Lattner's language — the creator of LLVM, Clang, and Swift — designed as a Python-syntax systems language that compiles native code for CPUs and GPUs. It hit a stable 1.0 the week before, freezing the API after three years of churn.

  • Qualcomm bought Modular — and opened it anyway

    The release lands right after Qualcomm acquired Modular, a move many expected to push Mojo proprietary. Going full Apache 2.0 instead removes the single-vendor risk that kept a lot of teams from betting on it.

  • A real escape hatch from CUDA

    Mojo targets GPUs and AI accelerators directly and does two-way Python interop, so you can call Python from Mojo and vice versa — a path to write kernels and inference without CUDA lock-in. macOS and Linux work today; Windows support is in progress.

  • The catch: no compiler PRs yet

    You can read, fork, and build the compiler now, but Modular won't accept compiler contributions until the end of 2026; only stdlib PRs are open, as they have been since 2024. Fine for learning and forking — not yet for upstreaming fixes.

  • Elsewhere: a 27B model that matches GPT-5.6 Luna

    Alibaba's open-weight Qwen3.8-27B scored 52 on the Artificial Analysis Intelligence Index, tying GPT-5.6 Luna while running frontier-class coding and reasoning locally on a single 24GB GPU — no cloud API required.

  • Elsewhere: Copilot retires 6 models Sept 1

    GitHub Copilot drops Gemini 3.1 Pro, Claude Opus 4.5 and 4.6, Sonnet 4.5 and 4.6, and Raptor Mini on September 1. Repin to Opus 4.7/4.8/5, Sonnet 5, or Gemini 3.6 Flash before then; Sonnet 4.6 survives only on annual individual plans.