Meta Launches Muse Code, an AI Coding Agent for Large Code Bases

My take: On August 5, Meta launched Muse Code, its first terminal coding agent, built by Meta Superintelligence Labs. The product competes directly with Claude Code, OpenAI's Codex CLI, and Google's Antigravity, and arrives with a pricing proposition that puts the decision at a specific point: $1.25 per million input tokens at its standard price, or $0.10 if the user allows Meta to use their source code to train its models.

That difference, twelve times cheaper in exchange for your code, is the part that requires analysis before deciding. For a personal or learning project it may make sense. For a team with proprietary code, the implications around confidentiality and intellectual property rights deserve a formal conversation before accepting the terms.

Competition in coding agents is good news for anyone working with code today: more options and pricing pressure lead to better tools over time. But the pattern of "lower price in exchange for training data" will repeat itself in this space. Does your team have a clear policy about what code can be sent to external AI platforms?