Meta Launches Muse Wonder, Its Most Capable AI Model Yet

Meta released two new models on October 12: Muse Wonder and Muse Wonder Full. Meta calls them its most capable so far, aimed at agent tasks where an AI works through a goal over many steps.

Two features are the headline. Active Memory lets the model keep and reuse what you tell it over time. Active Focus helps it stick with a hard or vague task until it is done.

Who can use them:

• Muse Wonder: available now through the Meta Model API and the Muse Code tool. TechCrunch reports API pricing of $4 per million input tokens and $20 per million output tokens.
• Muse Wonder Full: available now only to Muse Code subscribers, and not through the API.
• Wonder will also soon power the Muse assistant, starting with subscribers.

The catch: The performance claims come from Meta’s own testing. Outlets report different independent benchmark scores, so treat any ranking as unsettled for now. The strongest version is also limited to subscribers, so most developers will only see the standard one.

Try it: If you build with agents, test Muse Wonder on one of your own real tasks before switching anything.