Summary
Meta has introduced Muse Wonder and Muse Wonder Full, models designed for personal and professional tasks that can span multiple steps. Wonder is available through Meta Model API and Muse Code; Wonder Full is available to Muse Code subscribers.
Meta introduced Muse Wonder and Muse Wonder Full on October 12, 2026, describing them as its most capable models to date and building them for personal and professional tasks involving multiple steps. Meta says Muse Wonder is available to developers and builders through the Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers. Meta says Muse Wonder will also power more experiences in Muse, rolling out first to subscribers. The company’s announcement does not include numerical benchmark results alongside its performance claims.
Features aimed at tasks that unfold over time
Meta points to tasks such as navigating an insurance claim, understanding a lease and building an app as examples of what the models are intended to handle. It says the models are designed for work in areas including law, finance and health, as well as coding.
The company describes two features for carrying work across interactions. Active Memory is intended to retain relevant information between conversations and bring details back when they matter to a task. Active Focus is designed to keep pursuing a goal when a task becomes difficult, try different approaches, recover from errors and ask for clarification when needed. Meta says the models should check in before taking important follow-up actions.
These features are intended to support tasks that require continuity, rather than treating each exchange as an isolated request. The announcement describes their intended behaviour but does not detail how the features are implemented.
Safety and availability
Meta says its models are trained to avoid sharing private information without permission and to refuse requests that would assist harmful acts. The company also says it wants the models to help improve security and defend against cyberattacks. Developers and security teams can use them to look for flaws in their own code, so Meta says it has aimed to preserve that defensive capability.
The company says it trains, evaluates and releases its most capable models according to its Meta Superintelligence Scaling Framework. It says a fuller report with additional safety evaluations will be available soon.
For now, the two models have different access routes: Muse Wonder is offered through the API and Muse Code, while Muse Wonder Full is available to Muse Code subscribers. Meta says Wonder will reach the Muse service later, starting with subscribers, but the announcement gives no rollout date.