Meta Announces Muse Wonder Models for Agent Tasks

What Meta announced

Meta announced Muse Wonder and Muse Wonder Full on October 12, 2026, describing them as its most capable models to date. The company says the series builds on Muse Spark and is designed for personal and professional “agentic” work.

Meta cites tasks such as navigating an insurance claim, understanding a lease and making sense of finances. It also says the models are intended for expert fields including law, finance and health, as well as coding work for projects such as games or business apps.

Conversation continuity and task follow-through

According to Meta, the models include “Active Memory,” intended to retain relevant details across conversations. “Active Focus” is designed to keep pursuing a goal when work becomes difficult, try alternatives, recover from errors and ask for clarification when needed.

Meta says the models should check in before carrying out anything important. These are company-described capabilities, rather than independently reported product tests.

Safety claims and access

Because agents may work with messages, calendars and documents, Meta says it trained the Muse Wonder family not to share private information without permission, to refuse requests for harmful acts, and to support cyber-defense work. The company says a fuller safety-evaluation report will be available soon.

Muse Wonder—not specifically Muse Wonder Full—is available through the Meta Model API and Muse Code, Meta said. The company also said it will soon power more of Muse, first for subscribers; no timing or broader access details were provided.