Meta said on November 6 that specialized models with new capabilities are rolling out to Muse and Meta AI, with the current Muse model scoring 47% on Meta’s Personal Superintelligence Index, up from 38% for its predecessor.

The rollout extends a plan Meta outlined on October 12, when it described Muse Wonder and Muse Wonder Full as its most capable models and said Wonder would first support subscriber experiences before informing specialized models for all Muse users.

The deployed models are specialized versions of existing systems, improved through Muse Wonder rather than being the Wonder models themselves, and bring Active Memory and Active Focus to both assistants.

Active Memory is designed to retain important information, separate lasting details from temporary ones, and retrieve useful context in later conversations; Meta says the Muse models forget information critical to users’ goals 2.4 times less often and use critical context correctly in 90% of Active Memory Eval tasks, up from 76%.

Active Focus helps an assistant persist, recover or find alternatives when unexpected obstacles disrupt a task, and Meta reports a 67% improvement in Muse’s completion of everyday tasks involving unexpected issues.

More broadly, the models are built to multitask, collaborate with users during ongoing work and incorporate corrections without losing sight of the goal; in Meta AI, they can also break complex problems into steps, ask sharper follow-up questions and adapt when users change direction.