Meta's New Mac Coding Agent Costs Up to 20x Less If You Let Meta Train on Your Data

Meta today launched Muse Code, a terminal coding agent available for macOS and Linux. Muse Code is in beta, and it is powered by Muse Spark 1.2, Meta's newest model.

Meta CEO Mark Zuckerberg said Muse Code can accomplish complete software engineering tasks across large repos. It can plan out changes, write code, and validate results using multiple persistent subagents.

Muse Code is Meta's equivalent of OpenAI's Codex and Anthropic's Claude Code, and it is Meta's first major push into agentic coding.

There are several default skills available, such as /plan for turning a task into an approval-gated plan, /grill for stress testing a plan, and /goal for working toward the successful completion of an objective. Muse Code maintains an append-only local event log of every model call, tool run, approval, and edit, so it can resume where it stopped if it crashes.

Muse Code can be used through the Mac's terminal after a one-command install, and there is no dedicated app available. The model is priced at $1.25 for 1 million input tokens and $4.25 for 1 million output tokens, but Meta is offering a discount for users who give the company permission to use their prompts and completions to train future AI models.

With the Contributor tier enabled for Meta model training, Muse Spark 1.2 is priced at $0.10 for 1 million input tokens and $0.20 for 1 million output tokens. The Standard tier costs more and does not let Meta use prompts to train its models.

The more affordable Contributor tier is rate limited at 60 requests per minute and 2,100,000 tokens per minute, while the Standard tier allows for 3,000 requests per minute and 4,000,000 tokens per minute.