Meta announced specialized versions of its existing models on November 6, trained using Muse Wonder and rolling out to Muse and Meta AI users. The change brings selected capabilities from the frontier model family into the products people already use.
The update includes Active Memory, which is intended to identify when earlier information matters and distinguish durable facts from stale context. Meta reports a 2.4-fold reduction in forgetting critical information in its Muse evaluation. Its accompanying chart shows 90% versus 76% on correctly using critical context.
Active Focus is designed to help an agent recover when circumstances change or a task encounters a bottleneck. Meta reports a 67% improvement in completing everyday tasks involving unexpected problems. The post also describes better adaptation to user clarifications and a greater ability to switch between independent work and live collaboration.
These measurements are Meta’s own evaluations, and the announcement does not establish the same gains on every user’s work. The useful question is whether the assistants preserve the context that matters and respond well when a plan changes. The rollout is a product-model update, rather than a fresh release of downloadable Muse Wonder weights.