Meta has announced two new models — Muse Wonder and Muse Wonder Full. These are the company’s most capable models to date, designed to power personal and professional AI agents.
The models are built to handle practical tasks such as processing insurance claims, analyzing rental agreements, working with financial data, and coding from games to new business applications.
They feature two key capabilities. Active memory gives the models the capability of keeping vital user information during dialogues and recalling necessary data when appropriate for the task at hand. Moreover, active concentration ensures either continuity as the models solve complex problems, trying different ways to do it, correcting their mistakes, and asking the right questions when needed.
In addition, the models support agent commerce scenarios, performing commercial actions on behalf of the user, such as placing orders, making reservations, and processing payments within defined rules.
“Fourteen months ago, we began an overhaul of our AI efforts, and in that time we’ve gone from rebuilding our foundations to pushing the frontier,” Meta said in its announcement.
Muse Wonder is already available to developers through the Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers. Soon, Muse Wonder will power interactive applications within the Muse ecosystem, starting with subscribers.
Meta said the models were trained with a strong focus on safety. They are designed to avoid sharing sensitive information without user permission, refuse assistance with harmful activities, and help strengthen defenses against cyberattacks. The company stated that the training and release of the models follow its internal “meta-superintelligence” scaling standards. A full safety report will be published later.
The new models build on the previous Muse Spark lineup. The successful launch of the Muse personal AI agent last month could open a promising path for Meta to turn its substantial AI investments into new revenue streams through a subscription model, despite ongoing concerns about user data privacy.
