2 min read AI-generated

Meta Muse Code: A New Coding Agent Takes Aim at Claude Code

Copy article as Markdown

Meta unveiled Muse Code — an AI agent for large codebases that handles complete engineering tasks. It runs on Muse Spark 1.2 and fans big jobs out into parallel sub-agents. The pitch: cheaper than Codex and Claude Code.

Featured image for "Meta Muse Code: A New Coding Agent Takes Aim at Claude Code"

Meta has sent a new AI coding agent into the arena: Muse Code. According to Mark Zuckerberg, it handles “complete software engineering tasks across large repos” — planning changes, writing code, validating results. Currently in beta, and installable with a single command.

What Muse Code does

Under the hood runs Muse Spark 1.2, a new coding-focused version of Meta’s frontier model. Notably, the model was trained alongside the tool itself. So Meta didn’t just wrap an existing model in an agent — it tuned both together.

The most interesting part is how Muse Code deals with large projects. “When a job is big enough, it fans out to separate sub-agents working in parallel in isolated worktrees,” Zuckerberg explained. If you know Claude Code, that rings a bell — this exact pattern of parallel agents in isolated worktrees has been a big topic there for weeks.

The pitch is all about price

Meta positions Muse Code openly as a cheaper alternative to OpenAI’s Codex and Anthropic’s Claude Code. Alexandr Wang, Meta’s AI chief and head of the Superintelligence Labs, told the Wall Street Journal: “For a lot of workflows and a lot of use cases, this can be an incredibly good option, especially from a cost perspective.”

The numbers behind it: $1.25 per million input tokens, $4.25 per million output tokens, on a pay-as-you-go basis. That undercuts the established coding agents by a clear margin.

My take

For us Claude users, this isn’t bad news — quite the opposite. Competition keeps Anthropic on its toes. And the fact that Meta copies the worktree sub-agent pattern of all things shows that the approach Claude Code takes is becoming the standard.

The real question is quality. A cheap agent isn’t much use if it stumbles on complex refactors. Meta promises a lot, but Muse Spark 1.2 still has to prove itself in real, large codebases — exactly where Claude Code is strong. I’m curious about the first independent comparisons. Until then: price is an argument, but not the only one.


Sources: