Muse Wonder

Muse Wonder and Muse Wonder Full are AI models from Meta that help write software and carry out tasks with several steps. Meta released the family on 12 October 2026, with tools for coding, planning and computer use.

Both variants have a one-million-token context window, the amount of material a model can work with in a request. Their access differs: Wonder is available through the Meta Model API and Muse Code, while Full requires a Power Usage or Maximum Usage Muse Code subscription.

Choose the access route for the task

Meta’s product page lists these options as of 13 October 2026:

The API overview labels Wonder’s API a preview. Use the documented model identifier, muse-wonder-1.0, and check the chosen service’s current availability and limits. A subscription allowance and a token-based API bill measure usage differently.

Longer tasks still need checks

Meta describes the models as able to plan work, use tools and recover from errors during long tasks. It also describes active memory for retaining useful context. These are the company’s descriptions of the models’ intended capabilities.

A useful evaluation starts with a real task, a clear completion check and the permissions the task requires. For coding, check that the resulting software behaves correctly. For computer use, check the action and its result before granting access to more sensitive systems. The AI agents directory covers tools that put these capabilities to work.

Compare a benchmark with its test conditions

Meta publishes results across coding, computer use and professional tasks on its model page. Scores depend on the model version, reasoning setting and surrounding tools. Compare results from the same test and setup before drawing a conclusion about which model suits a job.

The LLM directory tracks other releases. Muse Spark remains a separate Meta family with its own versions and prices.