Meta Rolls Out Muse Wonder-Trained Models to All Users
Meta started rolling out models trained on its Muse Wonder frontier model to all Muse and Meta AI users on Friday, bringing Active Memory and Active Focus to free-tier users.
Meta (META) began rolling out new models trained on its Muse Wonder frontier model to all users of its Muse agent and Meta AI starting Friday, the company said, while keeping Wonder itself for harder subscriber tasks.
The new models are specialized versions of Meta's existing ones, trained on the behaviors that matter in a consumer app, the company said. They carry two features first introduced with Wonder: Active Memory, which carries details from earlier conversations into new ones, and Active Focus, which keeps models working when a task hits a bottleneck.
On Meta's internal evaluation, the new models correctly use critical context in 90% of tasks, up from 76% for the previous model. Meta also reports a 67% improvement in finishing routine tasks when the models run into unexpected problems, though it did not share the baseline for that figure.
Meta said it trained the models to recognize what information is temporary versus permanent, to reason across time, and to distinguish what is still relevant from what has become stale. The company has not named the new models or detailed its training method.
Muse, Meta's personal agent app, launched in September on the smaller Muse Spark model. Wonder and the more capable Wonder Full followed on Oct. 12. Wonder is available through the Meta Model API and in Muse Code, Meta's coding agent, while Wonder Full is limited to Muse Code's two most expensive plans. On the API, Wonder costs $4 per million input tokens and $20 per million output tokens. Muse has a free tier, and Meta AI runs inside WhatsApp, Instagram, Facebook and Messenger.
Meta has not disclosed whether the rollout is staged by region, account tier or device platform. Meta AI Research said the models distill behaviors from the larger frontier architecture into lighter systems suited to everyday consumer interactions, a likely cheaper way to reach that many users than serving Wonder itself.
The information comes from The New Stack's report on Meta's announcement.