Meta’s Claude Code rival exits beta with three new subscription tiers — and it’s pushing hard on price
The New Stack Paul Sawers ● Covered by 7 sources
Meta just launched Muse Code with $5, $15 and $50 monthly plans. It’s trying to beat Claude Code and Codex on price, fast, while adding SDK and workflow tools.
Based on reporting by The New Stack, Paul Sawers — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Meta has moved Muse Code out of beta less than a month after its debut, and the pitch is blunt: more features, more ways to buy it, and a lot less sticker shock than the competition. The coding agent, powered by Meta’s Muse Spark 1.2 model, is aimed squarely at Anthropic’s Claude Code and OpenAI’s Codex, with the same basic promise of helping with serious work across big codebases.
The launch version already handled long-running engineering jobs with asynchronous background agents. Meta has now added inter-session messaging, so separate sessions can share context when they touch the same work. It has also introduced Workflows, which coordinates specialized agents across bigger tasks and surfaces that process in a control room. There’s a rewind feature too, letting developers step back to an earlier point in the conversation and code.
Pricing was always going to be the real fight here. Meta’s original setup included a Standard pay-as-you-go tier at $1.25 per million input tokens and $4.25 per million output tokens, plus a much cheaper Contributor tier at $0.10 and $0.20. But Contributor required users to let their prompts and completions help improve Meta’s products, which made it a non-starter for some teams working on proprietary code.
Now there are subscriptions as well: Everyday Usage at $5 a month, High Usage at $15 for three times as much usage, and Power Usage at $50 for 10 times as much. Meta says the $5 plan usually gets 10 to 50 requests every five hours, depending on how hard the task is. That headline price undercuts Anthropic and OpenAI, whose coding agents sit inside broader $20, $100 and $200 plans, even if the actual amount of useful work is still messy to compare.
The New Stack’s own tests found that cheaper doesn’t automatically mean better value. Muse Code used far more tokens than Claude Code on the same three tasks, and its output was weaker on a refactoring job and left dead code behind on another. So the real question isn’t just what Meta charges. It’s how much useful code developers get for the money.
Meta is also widening the moat around Muse Code with a new SDK in developer preview and the Muse Session Protocol, which lets outside tools talk to sessions directly. That opens the door to IDE integrations, internal tools, and new agent products built on top of Muse Code. Less than a month after beta, Meta is already pushing hard to make Muse Code feel like infrastructure, not just a terminal tool.
My take — AI-written commentary, not fact-checked reporting
Meta’s playbook is obvious: race to the bottom on price, then dress it up as developer empowerment. That usually works until the bill comes due in data, lock-in, or both. Open-model fans should be watching closely, because cheap access is great right up until the business model starts whispering in your ear.
Read more about this at: The New Stack