What Are Muse Wonder and Muse Wonder Full and How Do They Work?

Meta has released two new AI models called Muse Wonder and Muse Wonder Full. They are designed to handle complex tasks that require multiple steps, like sorting out an insurance claim or writing code for an app.

The rebuild behind the models

Fourteen months ago Meta began an overhaul of its AI efforts. The company says it has gone from rebuilding foundations to pushing the frontier. Muse Wonder and Muse Wonder Full sit on groundwork laid with Muse Spark. The announcement calls them the most capable models to date. They target personal and professional agentic tasks. That phrase means the models can take a goal and work through the steps without constant supervision.

Active Memory and Active Focus

These systems carry two named capabilities. Active Memory lets them keep what matters across conversations and bring back details relevant to the task at hand. Active Focus lets them keep working toward a goal when a task gets hard. They try other approaches. They recover from errors. They follow through. They ask for clarification when needed. They check in before following through on anything important.

Real work unfolds over time. A lease review might span days. A coding project might span weeks. The memory feature aims to make that continuity possible. The focus feature aims to make the model persistent rather than brittle.

Expert domains and coding

The announcement lists law, finance, and health as expert domains. It also claims frontier coding performance for people building everything from a game to an app for a new business. No benchmark numbers appear in the text. No comparison scores against other models appear in the text. The research blog is promised for a closer look at how the models were built, how they were tested, and what they can do.

Safety training and release standards

Training focused on avoiding private information sharing without permission. The models turn down requests to assist in performing actually harmful acts. They work to improve security and strengthen defenses against cyber attacks. Developers and security teams can use the models to find flaws in their own code and strengthen their systems. The company says it made sure the models can still do that work.

Release follows the Meta Superintelligence Scaling Framework. A full report with additional safety evaluations will be available soon. The announcement does not say when.

Availability today and tomorrow

Muse Wonder is available today through the Meta Model API and Muse Code. Muse Wonder Full is available today for Muse Code subscribers. Muse Wonder will soon power more of Muse, rolling out first to subscribers. The announcement does not define what Muse is beyond calling it the world's first personal AI agent built for everyone in a previous September release. It does not say how many subscribers exist. It does not say what the rollout timeline looks like beyond "soon."

Where the explanation stops

The document does not disclose model sizes. It does not disclose training data composition. It does not disclose compute requirements. It does not disclose pricing for API access. It does not disclose whether the models run locally or only in the cloud. It does not disclose latency figures. It does not disclose supported context window lengths. The research blog may address some of these. Until it appears, the boundary of public knowledge ends at the features named above.