Meta launches Muse Code, its first AI coding agent, to challenge Anthropic and OpenAI

Meta has unveiled Muse Code, its first AI-powered coding agent, as the company intensifies its push to compete with leading AI labs Anthropic and OpenAI in the rapidly expanding developer tools market. The release marks a significant step in Meta's broader artificial intelligence strategy, which has come under renewed scrutiny following a recent dip in investor confidence.

Muse Code is the latest major product to emerge from Meta Superintelligence Labs, led by AI chief Alexandr Wang, who joined the company in June of last year as the centerpiece of CEO Mark Zuckerberg's effort to overhaul Meta's previously struggling AI initiatives. Wang described the new tool as a versatile solution for software engineering tasks, noting that developers can install it with a single command and use it to plan changes, write code, and validate results across a wide range of use cases.

The launch comes at a critical time for Meta, which continues to invest heavily in data centers and related computing infrastructure to support its AI ambitions. Last week, the company's shares tumbled after Meta issued a conservative revenue forecast and reported shrinking free cash flow in the second quarter, underscoring the financial pressure tied to its aggressive capital spending.

Muse Code joins a competitive field that includes Anthropic's Claude and OpenAI's Codex, both of which have gained traction among developers seeking to streamline application building within a unified interface. The tool also enables users to manage fleets of AI-powered digital agents, a capability that is becoming increasingly central to modern software development workflows.

The coding agent is available in a preview version and operates alongside Meta's latest AI model, Muse Spark 1.2. While Wang declined to disclose specific user statistics for the Muse Spark family, he said adoption has been "exciting and strong." He added that the newest model was developed and trained in tandem with Muse Code, which he said enhances overall coding performance.

Meta is positioning Muse Code on the basis of affordability rather than raw capability. Developers can access the tool through a pay-as-you-go option, and Wang highlighted a "contributor tier" that offers access at a significantly lower cost, which he said is "more than 10 times cheaper than even the pay-as-you-go tier." In exchange for the reduced price, developers must opt in to help improve the model by allowing Meta to use third-party data to refine the underlying technology.

Underpinning Muse Code is a so-called harness, a system that lets developers manage AI models tailored for coding projects. Users will be able to access and pay for the tool through the same Meta developer page that hosts the Muse Spark AI model API. Additionally, the new model will be available on OpenRouter, a platform that hosts popular AI models, including open-weight models from Chinese labs such as DeepSeek and Z.ai.

Wang also revealed that Meta is beginning to accept requests for zero-data retention, a feature designed to assure enterprise clients that their development data will not be used to improve models. He described the option as "a big enterprise feature that is important for folks." The move is notable given that Meta generates 98% of its revenue from online advertising, a business model that relies heavily on user data for targeted marketing.

The debut of Muse Code signals Meta's intent to carve out a meaningful share of the AI developer tools market, even as it navigates the financial realities of its infrastructure investments. By competing on price and enterprise-friendly features, Meta is seeking to differentiate itself in a crowded field dominated by well-funded rivals.