What changed
Meta introduced Muse Wonder and Muse Wonder Full on October 12, describing them as its most capable models for personal and professional agent tasks. The most useful way to assess that claim is to inspect the concrete tasks Meta chose to show, rather than repeat a benchmark headline.
On Meta’s research page, the insurance-claim demo follows a disputed medical bill; the marathon-planning demo uses a browser and map to assemble a route and plan around a person’s circumstances; the personalized-gifts demo turns Instagram photos into a small swag-order workflow; and the wedding demo builds a seating chart around several constraints. These are materially different jobs: document interpretation, web research, commerce and constraint management.
What the result actually shows
Meta attributes the behavior to Active Memory, which carries relevant details across conversations, and Active Focus, which retries, recovers from errors and asks questions when needed. The launch post says Muse Wonder is available to developers through Meta Model API and Muse Code; Muse Wonder Full is available to Muse Code subscribers. Consumer Muse integration is described as coming soon.
The demos make the product direction legible: an agent that keeps a goal alive across tools and steps. They do not establish general reliability across arbitrary accounts or prove every action can happen without approval. Meta says the models check before important actions, and a fuller safety report is still forthcoming.
Source published 2026-10-12. Coverage is based on the maker’s announcement and demonstration.