Meta just entered the AI coding agent market with Muse Code, now available in beta. The product is the first coding agent from Meta’s Superintelligence Labs unit, led by AI chief Alexandr Wang, who joined Meta last year to shore up the company’s frontier model efforts.
What Muse Code Does
Muse Code handles complete software engineering tasks across large codebases. Like Anthropic’s Claude and OpenAI’s Codex, it lets developers and non-technical users describe what they want to build in plain language and generates the code for them.
The agent runs on an updated version of Meta’s Muse Spark model, built specifically for this use case. When a task is large enough, it fans out to parallel sub-agents working in isolated worktrees. According to Zuckerberg, Meta tested it building six game features simultaneously with no collisions, and the user’s working copy is never touched during that process.
The Pricing Angle
Wang told CNBC that Meta is positioning Muse Code as the cheaper alternative to Claude and Codex. There are multiple pay-as-you-go tiers. The standout is a contributor tier that Wang described as more than 10 times cheaper than any other tier. The trade-off: developers on that tier must allow Meta to see their usage data, which Meta will use to improve Muse Code over time. Customers who want their data kept private pay a higher rate.
Where It’s Available
Muse Code and the underlying Muse Spark model are accessible via Meta’s APIs and through OpenRouter, the platform that hosts open-weight models from labs including DeepSeek and Moonshot AI.
The business context matters here. Meta’s shares dropped last week after an earnings report showing a light revenue forecast and shrinking free cash flow. Coding is one of the few areas where generative AI has already proven enterprise revenue at scale, and Muse Code looks like Meta’s direct play to capture some of that.
