On Nov. 6, 2026, Meta said it was rolling out specialized versions of existing models trained using Muse Wonder to users of its Muse agent and Meta AI. Meta says the models are designed for task-focused behavior, including remembering relevant context and adapting when a task changes.
What the rollout means for Muse
Meta says the models bring Active Memory and Active Focus to Muse. Active Memory is intended to retain relevant details from earlier interactions, distinguish temporary information from lasting context, and recognize when information has become stale. Active Focus is meant to help Muse continue toward a goal when an obstacle arises, by trying again, recovering or finding another approach.
Meta also says Muse can work independently or collaborate with the user, adapting to directions and clarifications while a task is underway. The rollout builds on Muse’s role as an agent that handles tasks, a subject of NeoTeo’s earlier coverage of Muse’s privacy and trust questions.
Meta describes a conversational role for Meta AI
For Meta AI, the company describes the models as supporting step-by-step problem solving, follow-up questions and adaptation when a conversation changes direction. That is a different emphasis from Muse’s task execution: Meta presents Meta AI as a conversational collaborator, while Muse can also work through tasks and respond to obstacles along the way.
Meta’s Muse results measure different things
Meta reported two separate comparisons between its current and previous Muse models. In its Active Memory evaluation, the current model correctly used critical context in 90% of tasks, compared with 76% for the previous model. On Meta’s separate Personal Superintelligence Index, the scores were 47% and 38%.
| Meta-reported measure | Current Muse model | Previous Muse model |
| Active Memory tasks correctly using critical context | 90% | 76% |
| Personal Superintelligence Index score | 47% | 38% |
Read the 90% and 76% figures as results for the critical-context evaluation; the Personal Superintelligence Index is a separate measure.
Meta also reported a 2.4× reduction in forgetting critical information for improved Muse models, and a 67% improvement in completing everyday tasks when unexpected issues arise. Meta’s announcement does not state a baseline for the 67% figure or provide full evaluation methodology for these results.
