Meta launches Muse Wonder for longer-running agent tasks

Meta’s Muse Wonder models target agents that retain relevant context across conversations and continue working through multi-step tasks. Meta calls these capabilities Active Memory and Active Focus. Muse Wonder is available through its model API and Muse Code; Wonder Full requires a Muse Code subscription. The launch’s performance comparisons are vendor evaluations, while rollout into the consumer Muse experience is still forthcoming.

Meta AI · Oct 12, 2026

Why it matters

Agent evaluations increasingly need to test remembered requirements and sustained progress, not just one-turn answers.

More context

Try a bounded repository task through the documented access paths and track completion, interventions and cost. Watch for the fuller safety report, which the announcement says will follow; do not infer open-weight availability from the launch.

Keep in mind

The announcement does not provide independent validation that the model will complete your long-running workflow reliably.

Original read