Meta has entered the AI coding-agent race with Muse Code, a terminal-based tool now available in beta for macOS and Linux. The release marks Meta's first major push into agentic coding, positioning the company directly against OpenAI's Codex and Anthropic's Claude Code.
The new agent is powered by Muse Spark 1.2, Meta's latest model that offers improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows. Unlike the prior version, Muse Spark 1.2 was developed and trained alongside Muse Code, which Meta's AI chief Alexandr Wang said improves overall coding performance.
Zuckerberg said in a social media post that Muse Code can accomplish complete software engineering tasks across large repositories, including planning changes, writing code, and validating results. For Mac users, installation is a single command, with no dedicated app available.
Muse Code handles large projects by launching its own agents that work simultaneously. When a job is big enough, it fans out to separate sub-agents working in parallel in isolated worktrees, with the developer's working copy never touched.
The tool maintains an append-only local event log of every model call, tool run, approval, and edit, allowing it to resume where it stopped if it crashes. Meta designed the agent to be handy at planning changes to a codebase, writing the code, and validating the results.
Developers can access Muse Code through pay-as-you-go pricing that matches the Muse Spark 1.1 API rates at $1.25 per million input tokens and $4.25 per million output tokens. Meta also plans a contributor tier that costs $0.10 per million input tokens and $0.20 per million output tokens, with users agreeing to provide feedback to improve the coding agent.
Wang described Muse Code as one of the most affordable coding agents on the market, available globally. The model is accessible through both Muse Code and the Meta Model API with expanded global access, though right now the tool and its model are only available on Meta's developer website and OpenRouter.
The release follows Meta's July launch of Muse Spark 1.1, which upgraded the model's agentic and multimodal abilities and introduced Meta's paid API service. Before that, Meta replaced Llama with the original Muse Spark in April, when the company admitted it needed to work more on the model's coding abilities.
Zuckerberg did not rule out opening up Muse Spark in the future, and the service offers a 1 million token context window with discounts available.
