Meta on Monday unveiled a new series of advanced flagship artificial intelligence models, as it tries to catch up to rivals like Anthropic and OpenAI in the AI race.

The two new models, called Muse Wonder and Muse Wonder Full, scored nearly as well as Anthropic’s most advanced models on tests that measure reasoning, writing and coding, according to data provided by Meta.

Mark Zuckerberg, Meta’s CEO, revamped his company’s AI division last year after setbacks in developing the technology. He has since made progress with new AI models and other AI products for the company’s 3.5 billion users across Instagram, Facebook and WhatsApp. Muse, Meta’s AI agent app that can operate software and apps on behalf of users, climbed to the top of the App Store in the weeks after it was released last month.

Muse Wonder arrives amid growing concerns over advanced AI models and agents, some of which have gone rogue, hacking into companies and meddling with government websites. Zuckerberg has tried to market Meta’s technology, which is built in a similar way to Anthropic’s and OpenAI’s models, with a different message than its rivals, saying that AI is meant to be useful and empowering, not dangerous.

Meta began developing Muse Wonder after hiring Alexandr Wang, a founder of AI startup Scale AI, to be its chief AI officer in June 2025. Muse Wonder is the second series of AI models released under Wang’s AI division and is the first that Meta has said is meant to compete with other cutting-edge “frontier” AI models, which typically require the most computing power to run.

Muse Wonder will soon power parts of Muse, the company said. Muse Wonder Full is the more powerful model and will be available only for users who subscribe to Meta’s premium AI coding tool, Muse Code.

The new models, which were internally referred to as Watermelon, are also significantly better at coding than Meta’s previous models. The AI coding market is dominated by Anthropic and OpenAI, though Meta has been trying to grow its business selling AI services to other companies.

(The New York Times has sued OpenAI and Microsoft, claiming copyright infringement of news content related to AI systems. The two companies have denied those claims.)

Muse Wonder faced some setbacks during development. In July, Meta paused a stage of AI development called “pretraining,” which delayed the release of the models until this month, the Times previously reported.

In a livestream on Instagram over the weekend, Zuckerberg teased the new AI models while barbecuing meats. Also on the menu, Zuckerberg said: “A big watermelon.”

—

This article originally appeared in The New York Times.

By Eli Tan/Gabriela Bhaskar
c. 2026 The New York Times Company