Meta has unveiled its latest AI models, Muse Wonder and Muse Wonder Full, taking aim at Anthropic and OpenAI, the New York Times reported on Sunday.

The two models delivered results close to Anthropic’s top model, based on Meta’s own data, in tests measuring reasoning, writing and coding capabilities, the report said.

Coding performance also improved sharply from Meta’s previous models. Meta calls the two models “Watermelon” internally, the New York Times said.

Muse Wonder will soon be deployed to some functions of Meta’s AI agent app, Muse, which operates software and apps on users’ behalf. Muse rose to No. 1 on the app store within weeks of its launch last month. The more powerful Muse Wonder Full model will be available only to subscribers of Meta’s premium AI coding tool, Muse Code.

Meta began development of Muse Wonder after hiring Scale AI founder Alexander Wang (알렉산더 왕) as chief AI officer in June 2025. Muse Wonder is the second model series released by an organisation led by Wang. It is also Meta’s first model that the company has said will compete with “frontier” models, the New York Times said.

Anthropic and OpenAI dominate the AI coding market. Against that backdrop, Meta appears to be expanding its business of selling AI services to other companies.

Concerns are growing as some AI models and agents have gone beyond control, including hacking companies and tampering with government websites. CEO Mark Zuckerberg said, “AI is not a dangerous being but a tool that is useful to people and empowers them,” and Meta is promoting its technology with a message different from rivals, the New York Times said.