Meta updates its Muse and Meta AI assistants to remember more and recover from snags

Meta says its Muse and Meta AI assistants should now forget less and handle surprises better. On Nov. 6, 2026, the company said it was rolling out specialized versions of existing models, trained using something it calls Muse Wonder, to people who use Muse and Meta AI.

What changed

Muse is Meta's agent, meaning an assistant that carries out tasks. Meta says it now has two new features. Active Memory is meant to keep relevant details from earlier interactions, tell temporary information from lasting context, and spot when something is out of date. Active Focus is meant to help Muse keep going toward a goal when it hits an obstacle, by trying again or finding another way.

Meta AI gets a different emphasis. Meta describes it as a conversational helper that solves problems step by step, asks follow-up questions and adjusts when a chat changes direction.

The numbers, and what is missing

• Meta says the new Muse model used critical context correctly in 90% of tasks, up from 76%.
• On its Personal Superintelligence Index, scores rose from 38% to 47%.
• Meta reports a 2.4× drop in forgetting critical information.
• Meta reports a 67% gain in finishing everyday tasks that hit unexpected issues.

These are Meta's own results. The company did not give a starting point for the 67% figure or lay out how it ran the tests. Until outside testing exists, the numbers show Meta's goals more than proof. Privacy and trust questions around Muse have been covered by NeoTeo before.