Meta introduces Muse Wonder: remembering a task and finishing it are different claims

Meta’s October 12 announcement separates Active Memory from Active Focus. Review retained context and completed work separately.

Meta introduced Muse Wonder and Muse Wonder Full for personal and professional agent tasks. Its research post describes Active Memory as retrieving relevant context across conversations, and Active Focus as persisting through difficult work, trying alternatives and checking in before important actions.

Ask two concrete questions

In an invented task, an assistant remembers that a report must be in Korean but fails to export it. The language requirement was retained; the deliverable was not completed. Conversely, producing a file in the wrong language does not satisfy the remembered requirement. These are separate checks, not a test of Muse.

A vendor description is a starting point

Our reading recommendation is to request evidence for the work you need: which constraints were retained, which steps ran and which result can be inspected. A capability name does not establish a success rate for your task. We did not independently benchmark either model.

Check the announced route to access

At announcement, Meta said Wonder was offered through the Meta Model API and Muse Code, and Full through Muse Code subscriptions. Muse experiences were described as forthcoming, starting with subscribers. This announcement does not verify access on your account today.

Published October 14, 2026. Research announcement: October 12; the newsroom page also shows an October 13 update. We read both official posts, without a model trial, subscription, API request or deployment of an agent.

Sources: Meta · Muse Wonder · Meta Newsroom.