At the Bloomberg Tech conference held in San Francisco on Thursday, June 4, 2026, Alexandr Wang, Meta's Chief AI Officer, announced the launch of Muse Code, the company's inaugural coding agent. This move positions Meta to compete with prominent AI firms such as Anthropic and OpenAI.
Muse Code, unveiled as a key project from Wang, who leads the Meta Superintelligence Labs focused on foundational model development, marks a significant step in CEO Mark Zuckerberg's initiative to revitalize the company's struggling AI endeavors. Wang joined Meta in June of the previous year, becoming instrumental in reshaping the company's artificial intelligence direction.
In a recent interview, Wang explained that Muse Code can be set up using a single command and handles a range of software engineering tasks, including project planning, code writing, and result validation. This innovative tool supports Zuckerberg's strategy to monetize AI capabilities amid substantial investments in data centers and computational infrastructure. The announcement comes on the heels of Meta's disappointing revenue forecast and a decline in free cash flow reported in the second quarter, causing the company's stock to drop.
Structured similarly to AI coding assistants from Anthropic and OpenAI, Muse Code simplifies app development through a unified interface while managing a suite of AI-driven digital agents that enhance the software development workflow.
Currently available in a preview version, Muse Code operates in conjunction with Meta's latest AI model, Muse Spark 1.2. While Wang did not disclose specific user metrics for the Muse Spark models, he noted that the adoption rate has been "exciting and strong." He emphasized that the recent Muse Spark iteration, developed alongside Muse Code, contributes significantly to coding performance.
Developers can utilize Muse Code via a pay-as-you-go model, which includes a "contributor tier" offering a much lower entry price point—more than ten times cheaper than the standard pay-as-you-go option. This cheapest tier requires developers to participate in enhancing the model by providing third-party data to improve the technology.
Wang mentioned that Meta’s pricing strategy distinguishes Muse Code and the Muse Spark model family from similar offerings by Anthropic and OpenAI, focusing on cost rather than capabilities.
The backbone of Muse Code is a "harness" that enables developers to manage AI models specifically designed for coding projects. Users can find and pay for Muse Code on the same Meta developer page housing the Muse Spark AI model API. Additionally, Meta's latest AI model will also be accessible through the OpenRouter platform, which features various well-known AI models, including those from Chinese firms like DeepSeek and Z.ai.
Furthermore, Wang noted that Meta is beginning to accept requests for "zero-data retention," indicating that the company will not store developer data for model improvement purposes. He described this feature as crucial for enterprise users. Currently, 98% of Meta's revenue stems from online advertising, a sector in which the company maintains a strong foothold by leveraging user data for targeted marketing.
