Mojo 1.0 lands: one stable language for CPU and GPU code
Modular froze Mojo's syntax with a 1.x stability promise, while OpenAI's SDK forces an httpx2 upgrade and Tencent turns text prompts into 3D worlds.
Copy markdown
Mojo 1.0 is here, and the syntax is frozen
Modular shipped Mojo 1.0 (release 26.5): var declarations, closures and a single Pointer type are unified, and the team commits to keeping 1.x changes 'primarily additive' so the language stops shifting under you. Install today with 'uv pip install --upgrade mojo'.
One source, from CPU to GPU, no CUDA
Mojo's pitch is writing high-performance kernels across CPUs, GPUs and accelerators from one codebase, without dropping to CUDA or C++. Paired with MAX ('uv pip install max[all]'), it targets AI inference and systems code where Python has always hit a wall.
The stdlib is open; the compiler is next
The standard library is open source, with ~200 contributors landing 1,100+ PRs across 200,000+ lines, and Modular reaffirmed it will open-source the Mojo compiler and toolchain in 2026. If you were waiting for the stability and openness signals before betting on it, this is them.
OpenAI's Python SDK v3.0.0 defaults to httpx2, and breaks pins
openai-python 3.0.0 makes HTTPX2 the default client and stops installing httpx automatically. If you pass custom transports/clients or pin httpx, you must migrate to HTTPX2 equivalents (a legacy escape hatch exists at runtime). Check your lockfile before you bump the version.
Tencent's WorldClaw builds explorable 3D worlds from a prompt
Hunyuan3D-WorldClaw generates large 3D scenes from text (or a concept image) as independently editable, textured meshes you can drop into a game engine and walk through. The paper and project page are live, built on Hunyuan3D, SAM3 and BlenderMCP, though weights aren't posted yet.