Meta introduced the Muse Wonder series — Muse Wonder and Muse Wonder Full — describing them as its most capable models to date, built on groundwork from Muse Spark and tuned for personal and professional agentic tasks across law, finance, health and coding. Meta said the models add Active Memory, which carries relevant context across conversations, and Active Focus, which keeps a model working toward a goal when a task gets hard, tries other approaches, recovers from errors and checks in before important actions. On safety, Meta said the models were trained to avoid sharing private information without permission, refuse genuinely harmful requests, and still support defensive security work; it pointed to its Meta Superintelligence Scaling Framework and promised a fuller safety-evaluation report soon. Muse Wonder is available via the Meta Model API and Muse Code and will roll out to Muse subscribers first.
Our takeMeta's pitch has quietly matured from "an agent for chores" to "personal superintelligence" — a big claim resting on two features, persistent memory and persistence of effort, that are exactly where agents usually fall down. The safety language is unusually specific for a launch post: refuse harm, do not leak private data, but keep helping defenders. Still, the release is long on positioning and short on evidence — the evaluations Meta says will arrive "soon" are what decide whether Muse Wonder is frontier or merely confident.