Meta announced specialized models trained using Muse Wonder for Muse and Meta AI on Nov. 6, 2026. In a post, AIatMeta described two capabilities, Active Memory and Active Focus, intended to help the models stay aligned with users’ goals as tasks unfold.
Active Memory helps recognize when earlier information matters and distinguish what remains relevant from what has become stale, according to Meta. Active Focus addresses a different problem: recovering from setbacks or finding alternatives when constraints change.
What Meta claims improves in Muse
For Muse, AIatMeta reported a “2.4x reduction in forgetting information critical to your goals” and a “67% improvement in getting unstuck as constraints evolve.” Meta described the practical benefit as less repetition for users and fewer dead ends when plans change.
Those figures apply to Muse in the announcement, not Meta AI. The post did not identify the baseline model, evaluation methods, sample size or tasks used to produce the measurements, so it offers limited detail for assessing the claimed gains.
What changes for Meta AI
For Meta AI, the company described benefits without numerical results. AIatMeta wrote that the improvements help it break complex problems into steps, ask sharper follow-up questions and adapt when users change direction.
The announcement described the models as coming to Muse and Meta AI, but did not specify regions, eligible account tiers, platforms or a detailed rollout schedule.
