Meta AI Research said on November 6, 2026 that specialized versions of the models behind Muse and Meta AI have entered rollout, with current Muse scoring 47% on the Personal Superintelligence Index against 38% for the previous model. The gain comes from training with Muse Wonder rather than deploying the Wonder series itself. The update adds Active Memory and Active Focus to both assistants. For business users, the test is continuity of work across long tasks.
Specialized models trained with Muse Wonder
The deployment covers specialized versions of Meta's existing systems, improved through Muse Wonder. Meta described Muse Wonder and Muse Wonder Full on October 12, 2026 as its most capable models to date. Wonder was first to power Muse experiences for subscribers, then to develop specialized models for everyone using Muse. The November rollout extends that plan from Muse to Meta AI, bringing capabilities introduced with Wonder to a wider audience without replacing the assistants' base models.
Active Memory allows the assistants to retain important information and retrieve it when it becomes relevant in a later conversation. Training covers the distinction between temporary and permanent information, reasoning over time, and recognizing when details stay relevant or turn stale. Selection and timing sit at the center of the design: the model must identify which earlier context applies now and which no longer fits. For improved Muse models, Meta reports a 2.4x reduction in forgetting information critical to user goals.
Active Focus targets persistence when an unexpected challenge interrupts work. Meta trained the models on varied bottlenecks designed to reflect real-world conditions, so they can persist, recover, or pursue alternatives for the user. In Muse, the company reports a 67% improvement in completing everyday tasks where unexpected issues arise. The models are also built to multitask, choose between independent progress and live collaboration, and accept steering without losing the task. In Meta AI, the same base is framed as clearer step breakdowns and sharper follow-up questions.
What longer memory means for business use
For companies, the change affects assistants used for extended work where context accumulates across sessions. Active Memory Eval shows the current model using critical context correctly in 90% of tasks, compared with 76% for the previous model. That figure concerns application of retained context, separate from the retention gain. In practice, the difference should appear in fewer repeated briefings, more stable handling of preferences, and tasks that survive interruptions. Small teams gain a cheaper way to keep history without manual notes, while large firms can run longer multi-step processes.
The limits center on verification of memory behavior in specific workflows. Meta's figures describe everyday tasks and controlled evaluations, not guarantees for regulated records, fast-changing project data, or conflicting instructions from several staff members. Buyers should check how temporary and permanent information is separated, how stale details are discarded, and how corrections during a task are applied. The update does not by itself mean the assistant suits autonomous execution, since persistence still depends on correct goal framing and access to tools.
The marker to watch is whether the 47% index score and the memory and focus metrics translate into stable releases inside Muse and Meta AI over the coming months. Sustained retention of critical context and recovery from interruptions in live use would confirm the specialized-model path. If scores stall or behavior varies between the two products, the Wonder-based training will look like an incremental step rather than a new baseline for assistants.
