Meta says Muse Wonder models are its most capable yet

Meta said Monday in a newsroom post that it is introducing Muse Wonder and Muse Wonder Full, which it calls its most capable models to date, built for personal and professional agent tasks.

Who can use them: Muse Wonder is available now to developers through the Meta Model API and Muse Code, and Muse Wonder Full is available now to Muse Code subscribers, Meta said. Muse Wonder will soon power more of Meta’s Muse personal agent, rolling out first to subscribers.

What they do: Meta says the models can handle tasks like navigating an insurance claim or making sense of a lease, bring expert-level ability in law, finance and health, and offer frontier coding performance. They have what Meta calls Active Memory, which carries relevant details across conversations, and Active Focus, which keeps working toward a goal when a task gets hard, tries other approaches and checks in before doing anything important.

Where they came from: Meta said it began overhauling its AI work 14 months ago and built the new models on groundwork from Muse Spark, its earlier model. The performance claims are Meta’s.

Safety: Meta says it trained the models not to share private information without permission and to refuse requests to help with harmful acts, while still letting security teams use them to find flaws in their own code.

What’s next: A full report with more safety evaluations will be available soon, Meta said.

Earlier coverage

• Meta opens Muse personal agent in the U.S. on iOS, Android (Sept. 8)

— Rowan Keel