Meta updates the Muse model and launches spaces and voice conversation features

On November 5 and 6, 2026, Meta made a series of announcements regarding updates to its AI assistant Muse. It began the phased rollout of models trained with Muse Wonder to Muse and Meta AI. Muse also started offering spaces, where users can create their own tools, and the voice conversation feature is being rolled out starting in the U.S.

Introducing models trained with Muse Wonder

On November 6, Meta began updating the models used in Muse and Meta AI. The newly introduced models are specialized versions created by training existing models with Muse Wonder, enhancing their ability to extract and utilize necessary information from past conversations and their ability to keep working even when they encounter problems.

View on Threads

Muse Wonder is a series of AI models announced by Meta on October 12 that handles complex tasks at work and in daily life [1]. At the time of the announcement, Meta explained that it would also be used to develop specialized models for Muse.

The specialized models incorporate Muse Wonder's Active Memory and Active Focus.

Models equipped with Active Memory can identify and utilize the information needed for the current task from past conversations. They are trained to distinguish between temporary information and information that remains unchanged for a long time, and between information that is still valid and information that has become outdated.

According to Meta's evaluation, the rate at which Muse forgets information necessary for achieving goals dropped to about one-2.4th compared with the pre-update model [2].

Models equipped with Active Focus can keep working even when they encounter unexpected problems, recovering or looking for alternative approaches. They are trained assuming various dead ends that can occur in real-world work. In Muse, the ability to complete everyday tasks to the end even when unexpected problems arise improved by 67%.

Meta says these improvements reduce the burden on users of having to repeat the same information and reduce situations where work stalls due to plan changes.

In addition, the specialized models were trained to handle multiple tasks and to decide whether to proceed with work on their own or while interacting with the user. Even when the user changes instructions or adds information during a task, Muse can now more easily adjust its approach accordingly.

Meta also stated that Meta AI has improved abilities such as "organizing the steps to solve complex problems," "asking more precise follow-up questions," and "responding to changes in direction during a conversation."

On the same day, Meta also announced spaces, a new feature of Muse. When users describe the tool they want in words, Muse creates it within the app. Users can create travel itineraries, gift management tools, daily word games, and more, and reuse them later.

After creation, users can also ask Muse to adjust the features and appearance. Created tools can be placed in the sidebar or main tab, and users can set when they are displayed. For example, users can put a weekly schedule in the foreground on weekdays and switch the display to weekend plans on Saturdays.

Also, by connecting apps users normally use to Muse, information within created tools is automatically updated. For example, a tool that gathers music information would reflect new concert dates. Users can also check off items within a tool or add notes, and their edits remain the next time they use it.

spaces is available in all regions where Muse is offered, and supports the mobile, web, and Mac apps.

View on Threads

Converse by voice, with an avatar reacting through expressions and gestures

Prior to this, on November 5, Meta began offering a voice conversation feature and a feature in which an avatar reacts with expressions and gestures in sync with that conversation. It is available in the mobile, web, and Mac apps, rolling out starting in the U.S.

In voice conversation, users can choose Muse's voice from hundreds of options. Users can organize ideas while speaking them aloud, plan trips, or practice for interviews. When users speak to request a task while taking a walk or cooking, Muse proceeds with that task in parallel with the conversation. Users can speak to Muse even while it is in the middle of responding.

When an avatar is displayed during a conversation, users can see not only Muse's expressions and gestures when it speaks, but also its reactions while listening to the user. The avatar can be displayed alongside the conversation screen or in full screen. Users can also converse by voice only without displaying it.

View on Threads