Breaking01 / AI
Meta offers Muse Wonder agent model on its API and Muse Code
Performance claims rest on company-published figures, and the announcement gives no API pricing or publication date for the full safety report.

Meta announced on 12 Oct 2026 that its agent model Muse Wonder is available through the Meta Model API and Muse Code. Muse Wonder Full, unveiled alongside it, is for Muse Code subscribers, according to the announcement.
Meta said Muse Wonder will soon come to its Muse service, starting with subscribers. It also plans to use the model to build specialised models for all Muse users.
Active Memory keeps a user's context across conversations and finds the details a task needs, according to the announcement. Active Focus tries other approaches when work is blocked and continues toward the goal. The model checks with the user before important actions.
Meta said company-published figures show state-of-the-art results in personal and professional agent, memory and coding evaluations. The announcement names Harvey's Legal Agent Benchmark, FocusBench and SWE-Marathon among the tests.
The comparison with the earlier Muse Spark 1.3 also rests on company-published figures. Meta said Muse Wonder Full exceeded the Muse Spark 1.3 trend prediction on tasks not targeted in post-training, the stage of refining a model after its base training. It attributed the gain to a larger, stronger base model.
Meta wrote in the announcement that "Muse Wonder is our most aligned model to date". It described the safeguards against chemical, biological, radiological and nuclear misuse and cyber misuse as the strongest of any Meta model. Cyber capabilities remain available for defensive use.
Meta said a full report with additional safety evaluations will come soon. The announcement gives neither the report's publication date nor API pricing.