Meta launched Muse Code in beta on 5 August 2026 — its first AI coding agent, and its clearest signal yet that it wants a seat at the table alongside Anthropic, OpenAI and Cursor rather than watching from the sidelines. Mark Zuckerberg announced it directly, and Meta’s AI chief Alexandr Wang was explicit about the angle: “this can be an incredibly good option, especially from a cost perspective.” Muse Code runs on Meta’s Muse Spark 1.2 model and charges $1.25 per million input tokens and $4.25 per million output tokens — undercutting the standard pricing of both Claude Code and Codex.
Functionally, it looks a lot like the terminal-native agents already on the market: install with a single command, hand it a task, and for large jobs it fans out into sub-agents working in parallel inside isolated worktrees, leaving your working copy untouched until the work is validated. Meta’s own demo showed it building six separate game features simultaneously without conflicts. There’s a pricing wrinkle worth knowing about before you install anything: a discounted “contributor” tier requires opting in to having your code used to train Meta’s models, alongside a full-price tier that doesn’t. For any team with client confidentiality obligations, that’s not a small print detail — it’s the deciding factor.
So what
A fourth credible AI coding agent from a company with Meta’s compute and distribution doesn’t change what your team should be building today, but it does confirm the pattern this site has been tracking all year: the AI coding tools market keeps fragmenting rather than consolidating, and price competition between vendors is only going one direction. If you’re choosing or re-evaluating your AI-assisted development stack, the model underneath and the data terms attached to it now matter as much as the interface on top. See how we approach AI-assisted development with that in mind, or get in touch to talk through what fits your team.