Meta has launched Muse Wonder and Muse Wonder Full, a new pair of artificial-intelligence models aimed at completing complex personal and professional tasks. The October 12 announcement marks a significant expansion of the company’s push into AI agents.

Meta calls the models its most capable yet. Its examples include interpreting documents and building software, with the emphasis on working through a task rather than simply producing a conversational answer.

What is available now

Muse Wonder launched through the Meta Model API and Muse Code. Muse Wonder Full became available to Muse Code subscribers.

The rollout to the consumer Muse agent is still ahead: Meta says Wonder will power additional Muse experiences soon, beginning with subscribers. The announcement does not mean every Muse user immediately has access to the new models.

Memory and sustained work

In its research explanation, Meta highlights two capabilities: Active Memory, which retrieves relevant information from earlier conversations, and Active Focus, which helps the system persist with difficult work, change approaches and recover from obstacles.

The company says the models were trained at its Prometheus data centre and evaluated on agent tasks, memory, coding and specialized work involving law, finance and health. Its examples include creating a playable game and arranging a wedding seating plan.

Those are company-reported capabilities. The Northern Current has not independently tested the models, and Meta’s claims of leading benchmark performance should not be read as proof that the systems will reliably complete every real-world assignment.

How Wonder fits into Muse

Muse is Meta’s personal AI agent. Its product page describes a system that can browse websites, create documents and work with connected apps, including email and calendars. Meta says it can continue tasks in the background after a user closes the app.

The distinction matters: Muse is the product people interact with; Wonder is a new model family intended to improve the intelligence behind some of that work.

Privacy and the next stage

Meta says it trained the Wonder family to protect private information and resist harmful requests, and that a fuller safety-evaluation report will follow.

For Canadian users and businesses considering AI agents, the practical questions are whether the tools can complete useful work consistently, what access they require and how consequential actions are approved. The next test for Wonder is how its launch claims translate into everyday use.

Sources: Meta’s October 12 launch announcement; Meta’s research explanation, October 12; Meta’s Muse product overview.

Lead image: existing TNC AI-generated editorial illustration of data-centre servers; not a photograph of Meta’s facilities or the launch.