Meta Platforms Inc. is getting more serious in its efforts to challenge leading artificial intelligence labs Anthropic PBC and OpenAI PBC with the release of its first AI coding agent, called Muse Code.

Announced in a blog post today, Muse Code is the brainchild of Meta’s AI chief Alexandr Wang, who heads up the company’s Superintelligence Labs unit and is tasked with leading its frontier model development. Wang joined Meta last year at a time when Meta Chief Executive Mark Zuckerberg realized the company’s AI efforts were falling behind those of its rivals.

Muse Code, available in beta now, is able to perform “complete software engineering tasks across large repos,” Zuckerberg said in a post on X.

Releasing Muse Code in beta today. It’s a terminal coding agent that takes on complete software engineering tasks across large repos: planning changes, writing code, validating the results. Powered by Muse Spark 1.2, a coding-focused model update. pic.twitter.com/xqavk41w6v

— Mark Zuckerberg (@finkd) August 5, 2026

Meta’s decision to develop a coding agent looks like an effort by the company to recoup some of the hundreds of billions of dollars it has been investing into AI data centers and the infrastructure that goes inside them. Coding has emerged as one of the most successful applications of generative AI, with thousands of enterprises globally now using the technology to supercharge their developer teams. For companies like Anthropic, it has become a major source of revenue, and Zuckerberg clearly believes it can be the same for Meta.

Meta needs to do something. Last week, its shares tumbled following a disappointing earnings report that saw it issue a light revenue forecast and dwindling free cash flow.

Like Anthropic’s Claude and OpenAI’s Codex tools, Muse Code is meant to make it easier for developers and non-technical users to build software applications. Rather than write every line of code by hand, users can simply describe the app they need to build in their natural language, and have the assistant generate the code for them.

Muse Code is powered by a newly updated version of Meta’s Muse Spark model. It controls fleets of AI agents that work in concert with one another to complete complex tasks. “When a job is big enough, it fans out to separate sub-agents working in parallel in isolated worktrees,” Zuckerberg explained. “Your working copy is never touched. In testing, we had it build six features for a game simultaneously with no collisions.”

The new version of Muse Spark was developed specifically to work with Muse Code, the company explained.

Wang told CNBC in an interview that Meta is trying to differentiate Muse Code by making it more affordable than either Claude or Codex. Developers can access the tool through various pay-as-you-go tiers, including a “contributor tier” that Wang said is “more than 10-times cheaper” than any other tier. However, to access this tier, developers are required to let Meta see what they’re doing with Muse Code. The idea is that Meta can use this third-party data to continuously improve Muse Code’s capabilities over time. Alternatively, customers can pay a higher rate to ensure that none of their data will be retained by Meta.

Muse Code is underpinned by what’s known in the industry as a harness, which developers can use to manage different AI models that are customized for specific coding projects.

Meta said it’s making Muse Code and the Muse Spark model available through its application programming interfaces and also the OpenRouter platform that hosts many of the top performing “open-weight” models made by Chinese AI labs such as DeepSeek Ltd. and Moonshot AI.

Image: Meta Platforms