Meta launches Muse Code, a terminal-based AI coding agent built to handle large software projects and compete with Claude Code and OpenAI Codex.

Meta has announced an early beta of Muse Code, a new coding agent meant to compete with Anthropic's Claude Code and OpenAI's Codex. The new terminal-based coding tool is powered by Muse Spark 1.2, a new version of Meta's AI model which offers "improvements in code generation,

Meta CEO Mark Zuckerberg has launched Muse Code, a new AI coding agent that he says can handle complete software engineering tasks from planning and coding to testing. The launch comes as AI leaders predict that traditional coding tools and even some software engineering work could soon be taken over by AI agents.

Meta co-trained its Muse Spark 1.2 model with the terminal-based Muse Code agent, but analysts say the approach does not yet set it apart from rivals.

Meta launches Muse Code AI coding agent to rival OpenAI Codex & Anthropic Claude Code—automates repo tasks with audit logs.

Right now, Muse Code and the model behind it are only available to developers on Meta’s developer website and OpenRouter, a site that hosts AI models. You’ll need an API key to access them. Anyone can technically buy an API key, but the technical nature of the launch means it’s likely best suited for software developers and engineers.

The default on-ramp for Muse Code sends developers' code and prompts into Meta's training pipeline — a tradeoff enterprises with proprietary codebases will need to consciously opt out of by moving to standard pricing.

Meta announced Muse Glimmer on Monday, a new open-weight AI model. Muse Glimmer is designed to run locally on "commercial" Macs and PCs, though the hardware demands are still high. Your computer should have anywhere from 24GB to 32GB of dedicated RAM in order to run Muse Glimmer locally.

Meta launches Muse Code, a new AI coding agent aimed at OpenAI and Anthropic, with aggressive pricing and a new path toward enterprise AI revenue.

Glimmer stakes out different ground: a dense model with native vision input, trained end-to-end around the agent loop, shipping with its own quantized variants and speculative-decoding drafter.