Meta Brings Wonder-Trained Specialized Models to Muse and Meta AI

What happened

Meta's AI research team said in its official blog that the Muse Wonder series, released a few weeks ago, is Meta's most capable model family to date, built for personal and professional agent tasks; the team used it to train specialized versions of its existing models and has begun rolling them out to people who use Muse and Meta AI.

Active Memory lets the assistant remember what matters and recognize when earlier information becomes relevant again; the models are trained to tell temporary information from permanent information, to reason across time, and to distinguish what is still relevant from what has become stale. Meta said that in Muse, the improved models show a 2.4x reduction in forgetting critical information needed to help users reach their goals.

Active Focus lets the assistant keep working toward a goal even when unexpected challenges arise; the models are trained on a variety of real-world bottlenecks so they try harder, recover, or find alternatives. Meta said that in Muse, everyday tasks show a 67% improvement in completion when unexpected issues arise; in Meta AI, the same improvements make for a better thinking partner that breaks complex problems into clear steps, asks sharper follow-up questions, and keeps up when users change direction mid-conversation. The same day, Muse also launched Spaces: users describe an idea in plain language and Muse builds a workspace that can be pinned to the sidebar or main navigation and stays updated from connected apps.

Key facts

1. Announced by: Meta's AI research team, official blog, Nov 6, 2026
2. Rollout: specialized versions of existing models, trained using Muse Wonder, for Muse and Meta AI users
3. Active Memory: remembers what matters, distinguishes temporary from permanent information, reasons across time, and tells stale content apart; forgetting of critical information in Muse is down 2.4x (Meta's own evaluation)
4. Active Focus: keeps working toward a goal through unexpected challenges, trying harder, recovering, or finding alternatives; everyday task completion in Muse is up 67% when unexpected issues arise (Meta's own evaluation)
5. Multitasking and adaptation: the models learned when to push on goals independently or collaborate live, and to adapt to steers and clarifications mid-task without confusion
6. Meta AI: the same improvements make for a better thinking partner that breaks complex problems into clear steps, asks sharper follow-up questions, and keeps up when users change direction
7. Spaces (launched the same day): users describe an idea in plain language and Muse builds a workspace such as a tracker, planner, or project board, pinnable to the sidebar or main navigation, schedulable by day, and updated from connected apps; sharing with other people is not yet supported
8. Background: the Muse Wonder series was released on Oct 12 as Meta's most capable model family to date; the Muse assistant launched in September, and Meta reportedly said on its October earnings call that it had reached 10 million users in the US and Canada
9. Figures: the 2.4x and 67% figures are Meta's own evaluations, with no independent validation yet
10. Rollout pace: the blog says only that the rollout has begun, with no completion timetable; Spaces is available wherever Muse is available