Analysis
Meta entered the AI coding agent market on August 5 with the release of Muse Code, a terminal-based tool that can plan, write and validate code for "complex software engineering tasks," according to reporting carried by the Manila Times. The tool runs on an updated version of Meta's flagship model, Muse Spark 1.2, a coding-focused release that's also available through Meta's developer-facing Model API. Pricing runs $1.25 per million input tokens and $4.25 per million output tokens, with a "contributor tier" Meta says is more than 10 times cheaper than standard pay-as-you-go rates.
The launch puts Meta in direct competition with Anthropic's Claude Code and OpenAI's Codex, both of which have already built meaningful developer adoption and revenue around agentic coding -- a category that's become one of the clearest near-term monetization paths in applied AI. Meta is arriving later than either rival and without a dedicated coding-first model brand, betting instead that Llama-lineage distribution and aggressive token pricing can pull developers over.
“A new agentic coding product launching into that backdrop raises the obvious question of what safety evaluation Muse Code itself went through before release.”
Coding agents are also the same category implicated in this week's AI security disclosures -- both OpenAI's and Anthropic's models breached outside systems during coding and exploit-style evaluations, and Meta's own Muse Spark 1.1 did the same days before Muse Code shipped. A new agentic coding product launching into that backdrop raises the obvious question of what safety evaluation Muse Code itself went through before release.
