AIBRIEF
Meta's new terminal AI coding agent, in beta this week, challenges OpenAI's Codex and Anthropic's Claude Code on cost for full engineering work on large repos.
Meta released Muse Code in beta this week as a terminal AI coding agent for large codebases. Alexandr Wang, who runs Meta's Superintelligence Labs, is positioning the product against OpenAI's Codex and Anthropic's Claude Code on one explicit lever: cost.
Wang's framing, relayed through the Wall Street Journal, is that Muse Code is "an incredibly good option, especially from a cost perspective." The product handles planning, code writing, and validation across large repositories by fanning work out to parallel sub-agents in isolated worktrees, leaving the developer's local copy untouched. Mark Zuckerberg said on X that internal testing had Muse Code "build six features for a game simultaneously with no collisions," an illustrative anecdote rather than a verified benchmark.
Meta's pitch is that the tool can run a full software engineering job on a big repo at a lower price than the established leaders, a wedge aimed at the cost-sensitive end of the market, not the benchmark-sensitive end.
Pricing details and independent benchmark comparisons remain undisclosed. The competitive test is whether developers already running Codex or Claude Code will switch on price alone.