Meta Launches Muse Code to Let AI Swarms Conquer Massive Codebases
Meta is coming straight for Anthropic and OpenAI with a new terminal beast that doesn't just autocomplete syntax, but unleashes an army of sub-agents to rewrite entire architectures in parallel.
The terminal-based tool runs on top of the newly released Muse Spark model and installs with a single command. Instead of staring at an individual script like a confused intern, Muse Code handles complex projects by spawning independent sub-agents that work simultaneously across isolated working trees.
According to Mark Zuckerberg, this compartmentalized setup prevents code conflicts entirely, citing an internal benchmark where the system authored six distinct game mechanics at once without corrupting the local working directory.
To undercut rival products like Claude Code and Codex, Alexandr Wang and the engineering team set pricing at $1.25 per million input tokens and $4.25 for outputs, throwing in extra discounts for teams willing to trade their prompts for future model training fodder.
The move directly follows the open-weights release of the 30-billion-parameter Muse Glimmer model under an Apache 2.0 license on Hugging Face, signaling a deliberate push deep into the enterprise developer ecosystem.
The race to replace junior engineers has officially escalated from polite browser chats into automated, multi-threaded terminal swarms. When tech conglomerates start discounting compute in exchange for peering into private codebases, automation gets significantly cheaper at the expense of corporate privacy.
Source: TechCrunch
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.