Meta rolls specialized Muse Wonder-trained models into Muse and Meta AI

Meta on 6 November 2026 said it is introducing specialized versions of its existing models, trained using Muse Wonder, and rolling them out to people who use Muse and Meta AI. The company is not shipping Wonder itself as the everyday model; it is using Wonder to train the models behind those products.

Active Memory and Active Focus

Active Memory is meant to help the assistants remember what matters, tell temporary information from permanent information, reason across time and recognize when earlier details have gone stale. Meta reports a 2.4× reduction in Muse forgetting information needed to help users reach their goals. On its Active Memory evaluation, the current Muse model correctly used critical context on 90% of tasks, compared with 76% for the previous Muse model.

Active Focus is designed to keep Muse working toward a goal when unexpected problems arise. Meta says it trained the models on real-world bottlenecks so they can try harder, recover or find another route, and reports a 67% improvement in Muse at completing everyday tasks where unexpected issues come up. The specialized models are also described as better at deciding when to pursue a goal independently or collaborate live with the user, and at taking mid-task corrections. For Meta AI, Meta describes clearer step-by-step help and follow-up questions, without separate numerical scores.

On Meta’s Personal Superintelligence Index, the current Muse model scored 47%, up from 38% previously.

Limits

The figures are Meta’s own evaluations; the post does not give sample sizes, protocols or the baseline completion rate behind the 67% claim, and it reports numerical results for Muse rather than a separate Meta AI score. Independent verification is not reported here. Muse Wonder itself remains the frontier series introduced weeks earlier.