Key Takeaways
- AI Launch: Meta unveils two AI models to rival OpenAI and Anthropic.
- Leadership Move: Zuckerberg strengthens Meta’s AI strategy under Alexandr Wang.
- Business Growth: Meta targets paid coding tools and enterprise AI services.
Meta unveiled Meta’s new AI models on Monday to strengthen its position in the AI market and compete with OpenAI and Anthropic in reasoning, writing, and coding.
Meta Targets Rivals With New AI Models
The new models, Muse Wonder and Muse Wonder Full, performed nearly as well as Anthropic’s most advanced models on selected reasoning, writing, and coding tests, according to data provided by Meta. The company did not establish that the models outperform its competitors across all tasks.
Muse Wonder is expected to power parts of Meta’s Muse AI assistant, while Muse Wonder Full will be available to subscribers of Muse Code, the company’s premium AI coding tool. The launch signals Meta’s effort to expand its AI products for consumers and developers while competing in the growing market for AI-powered software.
Meta’s AI products could reach a large audience through its existing platforms, including Facebook, Instagram and WhatsApp, which together serve billions of users. The company is also seeking to expand its business selling AI services to other organizations.
Zuckerberg Reshapes Meta’s AI Strategy
Meta CEO Mark Zuckerberg reorganized the company’s AI division after earlier setbacks in model development. In June 2025, Meta hired Alexandr Wang, co-founder of AI data company Scale AI, as its chief AI officer to help advance its AI strategy.
Muse Wonder is the second series of models released under Wang’s division and the first that Meta has explicitly positioned to compete with leading frontier AI models. These systems typically require substantial computing resources and are designed to handle complex tasks, including coding and reasoning.
The models were internally known by the codename Watermelon during development. Their release follows delays reported earlier this year, including a pause in a pre-training stage in July that pushed back the launch timeline.
Zuckerberg also promoted the models in an Instagram livestream over the weekend while barbecuing. He joked about “a big watermelon,” referencing the internal codename associated with the models.
Meta Seeks Growth in AI Coding and Business Services
The launch comes as AI coding tools become an increasingly important market for technology companies. OpenAI and Anthropic have established products in the sector, while Meta is working to expand its own coding and enterprise AI offerings.
Muse Wonder Full, the more powerful of the two models, will be restricted to Muse Code subscribers, allowing Meta to connect advanced model access with a paid product. Muse Wonder will be integrated into parts of the broader Muse assistant, potentially increasing its use among consumers.
Meta has also promoted a message that AI should be useful and empowering, distinguishing its approach from competitors amid growing concerns about the risks of advanced AI systems and autonomous agents. Some AI agents have demonstrated the ability to perform actions across software and online services, raising questions about how such systems should be monitored and controlled.
The company’s reported benchmark results offer an early indication of its progress, but independent testing will be key to assessing performance on real-world tasks. Pricing, broader availability and performance under different workloads will also influence whether the new models can attract developers and business customers.
Meta’s latest release marks another step in its effort to compete with leading AI developers and turn its research investments into products that support its advertising, consumer and enterprise businesses.
