Meta introduces Muse Wonder models for longer agent tasks
The launch adds developer access while putting memory and sustained task execution at the center of Meta’s model pitch.
THE STORY IN THREE POINTS
- Meta announced two Muse Wonder models on October 12.
- The launch separates API access from Muse Code subscription access.
- Reported capability results are attributed to Meta.
What was announced
Meta introduced Muse Wonder and Muse Wonder Full on October 12. The company describes models built for personal and professional agents, with coding among their uses. Muse Wonder is available through the Meta Model API and Muse Code; the Full model is available to Muse Code subscribers. These are the access routes stated in the announcement.
The details that matter
Meta highlights Active Memory for carrying relevant information between conversations and Active Focus for persisting through difficult tasks. It says Wonder will reach Muse experiences later, beginning with subscribers. That planned rollout should be distinguished from the developer access announced at launch.
Editorial context
For someone delegating work, persistence is useful only when the goal and stopping conditions remain clear. A practical evaluation might ask an agent to revise a small application, document its decisions, and show that the result works after several interruptions. Reviewers could then inspect both completion and unnecessary changes. Memory deserves a separate test: whether the agent retrieves relevant preferences without importing unrelated personal details. Those are editorial evaluation ideas, rather than findings from our own tests of the new models.
Sources: Meta: Introducing Muse Wonder and Muse Wonder Full: Frontier Models for Agents ↗
