Meta brings Muse Wonder‑trained models to Muse and Meta AI

Meta's move builds on the Muse Wonder series, which the company introduced a few weeks ago as its most capable models for personal and professional agentic tasks. By training specialized versions of its existing models with Muse Wonder, Meta is aiming to carry some of those frontier capabilities into the models behind Muse and Meta AI, reaching a broader set of users rather than only those with access to the top-tier Wonder models.

The reported gains are concrete: the current Muse model scores 47% on the Personal Superintelligence Index, up from 38% for the previous model, and shows a 2.4x reduction in forgetting critical information. In Muse, the models also show a 67% improvement in completing everyday tasks where unexpected issues arise. These figures suggest the specialized models are designed to be more reliable at retaining context and working through obstacles, though the body does not detail how these measurements were conducted.

Beyond the scores, Meta says the specialized models learned to be better multi-taskers, knowing when to push on goals independently or collaborate live with users, and to adapt to mid-task steers and clarifications without getting confused. In Meta AI, the same improvements are positioned to make it a better thinking partner that can break complex problems into steps and ask sharper follow-up questions. The body does not specify a release date or pricing for these specialized models.