Meta launches Muse Code with aggressive $5–$50 subscription pricing to challenge rivals

Meta has moved Muse Code out of beta and into general availability, introducing three monthly subscription tiers that position the coding agent as a significantly cheaper alternative to Anthropic's Claude Code and OpenAI's Codex.

The three plans—Everyday Usage at $5, High Usage at $15, and Power Usage at $50—represent a deliberate undercutting strategy. For comparison, Anthropic bundles Claude Code with its $20-per-month Pro plan, while OpenAI includes Codex in its $20 Plus tier. Meta's entry-level offering costs a quarter of the price.

New capabilities fuel the public release#

Beyond pricing, Meta has added several features designed to expand what developers can accomplish with Muse Code, which runs on Meta's Muse Spark 1.2 model.

Inter-session messaging allows separate coding sessions to share context and state when their work overlaps, streamlining collaborative workflows. A new Workflows feature orchestrates teams of specialized agents on larger jobs, passing intermediate results between stages and delivering a single consolidated output.

Developers also gain a rewind capability, enabling them to roll back conversations and code changes to earlier points in a session—useful for experimenting with different approaches without starting over.

The price question: simplicity vs. substance#

Meta's pricing structure stands out for its proportional transparency: each tier increase maps directly to advertised usage multipliers. The Everyday Usage plan typically allows 10 to 50 requests per 5-hour window, with High Usage offering three times that capacity and Power Usage providing 10x.

Engineering leaders have praised this clarity. Muhammad Navaid, a senior machine learning engineer, highlighted the straightforwardness on LinkedIn, noting that Meta's approach contrasts with what he calls the "usage-limit drama" surrounding Claude Code and ChatGPT pricing.

However, the real cost of work remains harder to evaluate. Lower token prices don't automatically mean lower operational costs if the tool requires more tokens to produce usable results. The New Stack's previous testing found that while Muse Code consumed substantially more tokens than Claude Code on comparable tasks, it also delivered weaker outputs in some cases—raising questions about the true cost-per-completed-job.

Some observers also flagged concerns beyond pricing. Ashok Gelal, co-founder of Msty AI, called the rates "unbelievably low" but emphasized that —a reference to the original Contributor tier offering discounted rates in exchange for allowing Meta to use prompts and completions for product improvement.