Meta Drops Muse Code — And Zuckerberg’s Coming for the AI Coding Crown on the Cheap

Mark Zuckerberg just tossed another fighter into the AI cage — and this one writes code without draining your bank account.

Meta officially unveiled Muse Code on Wednesday, a new AI coding agent designed to build software, repair bugs and handle engineering work without forcing users to manually type every line of code.

Translation: You tell it what you want built, and the robot gets to work while you sit back and pretend you understand what all those colorful lines in the terminal mean.

According to The Wall Street Journal, Meta is pitching Muse Code as a lower-cost rival to heavy hitters including Anthropic’s Claude Code and OpenAI’s Codex.

Meta Is Starting a Coding Price War

Here’s where things get spicy: Meta isn’t just trying to build another coding assistant — it’s trying to undercut practically everybody in the building.

Muse Code reportedly comes with two pricing levels. The standard tier is priced similarly to Meta’s existing Muse Spark model, while a cheaper feedback tier costs just 20 cents per million output tokens.

Yes, you read that correctly. Twenty cents.

The catch? Customers using the bargain-bin tier have to agree to provide feedback that Meta can use to improve the agent. So Zuckerberg gives you dirt-cheap code, and you help train his next digital genius. Everybody scratches everybody’s back.

For comparison, competing coding models can cost anywhere from roughly $10 to $30 per million output tokens once customers move beyond bundled subscription limits. That means Meta isn’t gently knocking on the competition’s door — it’s arriving with a battering ram marked “discount AI.”

Muse Spark Laid the Groundwork

Muse Code arrives after Meta rolled out Muse Spark 1.1 in July, an upgraded multimodal reasoning model built for coding, tool use, computer control and complicated multi-step assignments.

Meta’s official Muse Spark 1.1 announcement claimed major improvements in diagnosing difficult software bugs, adding features to large codebases and carrying out major code migrations.

The company also says Muse Spark 1.1 scored 61.5% on SWE-Bench Pro, a benchmark focused on solving real-world software-engineering problems. That result placed it ahead of several major competitors on the published leaderboard — although benchmark scores should always be taken with a grain of silicon, especially when companies are promoting their own machines.

It is also important to separate the two products: Muse Spark 1.1 is the underlying general-purpose AI model Meta has been promoting to developers, while Muse Code is the newly announced coding-agent product built specifically around software-development workflows.

Meta Employees Are Already Using It

Meta AI chief Alexandr Wang says some employees inside the company are already using Muse Code, with internal adoption expected to increase.

That matters because AI coding agents aren’t just fancy autocomplete tools anymore. These systems can inspect repositories, create entire projects, run commands, hunt down errors and make changes across multiple files with limited human involvement.

Anthropic scored an early hit with Claude Code, while OpenAI has been aggressively expanding Codex. Now Meta wants a piece of that action — and its weapon of choice appears to be brutally low pricing.

Zuckerberg Needs an AI Win

The release comes as investors continue pressing Meta to prove that its monster AI spending spree can actually produce serious revenue.

The company has poured enormous amounts of cash into data centers, computing chips and expensive AI talent. Zuckerberg reorganized Meta’s artificial-intelligence operation under Meta Superintelligence Labs, placing Wang in charge and recruiting researchers with compensation packages reportedly worth millions.

Now the bill is sitting on the table, and Wall Street wants to know who’s paying it.

Muse Code could be part of the answer. If Meta can attract developers with extremely cheap pricing, it could quickly build a large user base, gather valuable feedback and wedge itself between OpenAI and Anthropic in one of the hottest corners of the AI business.

Bottom line: Meta may not have invented the AI coding-agent party, but Zuckerberg just showed up carrying the cheapest bottle in the room — and suddenly everybody’s looking his way.