Meta Debuts AI coding Agent Muse Code

META has released Muse Code, its first AI coding agent, in public beta for macOS and Linux [doc2, doc3]. Built under AI chief Alexandr Wang and powered by the new Muse Spark 1.2 model, the agent is aimed squarely at Anthropic's Claude Code and OpenAI's Codex in what has become the most commercially proven category of frontier-model usage.

The product choices are deliberately narrow. Muse Code ships with no dedicated application and runs entirely from the terminal, installing with a single command and connecting directly to the Meta Model API. It targets large repositories, planning changes, writing code and validating results, and it keeps a local append-only event log of every model call, tool run, approval and edit, which makes the runtime replay-exact and restart-safe after a crash. Pricing is where Meta is pushing hardest: standard pay-as-you-go API rates, plus a contributor tier that cuts cost by more than 90% in exchange for data sharing.

That contributor tier is the strategically interesting part, because it converts a price war into a training-data acquisition channel and is difficult for a pure-play model vendor with enterprise confidentiality commitments to match. For Meta the near-term revenue is immaterial against a business still driven by advertising, so the relevant read is on the capex thesis rather than the P&L: developer tooling is a distribution wedge for the Meta Model API. What to watch is enterprise adoption despite the data-sharing terms, and whether Anthropic and OpenAI respond on price or hold on capability.

Related Stocks

• META