Muse Code Leaves Beta and Puts a Price Floor Under Coding Agents

Meta just pushed its coding agent out of beta. Muse Code is now a full product, barely a month after it first appeared. It is Meta’s answer to Claude Code and Codex, and it runs on a model Meta built itself.

This is not a small release tucked into a changelog. It brings new features, an SDK, and a pricing ladder. Developers who want to try it can start for five dollars a month and climb from there.

What Changed

Muse Code is a terminal-based coding agent. It lives inside a shell. You describe a task and it works across files, runs commands, and gets a codebase building. That shape is familiar to anyone who has used Claude Code or Codex.

The default engine is Muse Spark 1.2. Meta trains that model in-house, so it owns the stack from the weights up to the interface. Owning the model is how the company keeps cost down and ships something that undercuts the incumbents.

General availability is the headline, but pricing is the real story. Meta introduced three subscription plans. They start at $5 a month and run to $50. An SDK also ships, which means the agent can be wired into other tools instead of staying trapped in one terminal window.

The SDK matters more than it looks at first glance. It turns a single agent into a component that developers can call from their own scripts, hooks, and CI jobs. That is how a coding agent escapes the demo stage and becomes a real part of a build pipeline.

The Price Floor Just Dropped

Coding agents have been selling at a premium. Claude Code and Codex carry subscriptions priced well above what Meta asks. Muse Code has always been the cheaper option per request, and it was already narrowing that lead.

Now Meta added the thing the market kept asking for. Predictable monthly billing. Instead of watching a meter tick, a developer pays a flat fee. That removes the anxiety that comes with usage-based pricing and makes the agent easy to budget.

The gap is not small. A $5 entry tier sits closer to a hobbyist’s budget than an enterprise line item. Meta is clearly willing to give up margin to win the seat. For a company that runs on its own models, that trade makes sense. Every developer who picks Muse Code is a data point the model can learn from.

What Wins in the Agent Market

The takeaway is that the coding agent race has become a price war. When three capable products can do the same job, the low-cost provider with a decent model usually wins the wide middle. Anthropic and OpenAI built their lead on capability. Meta attacks from the other side, on cost and on access.

That puts pressure on the incumbents. Claude Code and Codex are strong, but they charge more. If Muse Spark 1.2 handles the everyday work of wiring up a repository and running tests, plenty of teams will not pay double the price for marginal gains. The premium has to be earned by reliability and depth, not by habit.

The next move belongs to Anthropic and OpenAI. They can cut prices, add monthly tiers of their own, or lean on features a cheaper rival cannot copy. History says price wins handshakes while depth wins renewals.

Pricing pressure also resets expectations across the category. Once a five dollar agent exists, customers treat a twenty dollar agent as a costly choice. That dynamic shows up in every software market, and it is about to show up in coding agents too.

For developers, the short version is that the market just got more interesting. Try the cheap agent. Compare it against the expensive ones. Let the difference decide. Meta shipped a real product with a real price. Everyone else now has to respond.

Agent infrastructure is the development layer of the AI Infrastructure report, which covers how a model becomes a service with an owner, an evaluation gate and a cost line.

Related reading. OpenAI Just Made the Agent Harness a Managed Service. A Full VMware Exit Costs More Than Three Years of Broadcom’s Bill. Dynatrace Davis AI Stops Chasing Alerts and Starts Finding Causes. AI Infrastructure Runs on Four Layers. Most Break Below the Model..