This is the 'Muse Wonder' series developed by Meta for personal and professional AI agents. The models unveiled on October 12, 2026, consist of two types: Muse Wonder and Muse Wonder Pool. Meta introduced these as its highest-performing models. Their primary uses are multi-step tasks and coding.<\/p>\n\n\n\n
The core features are 'Active Memory' and 'Active Focus'. Active Memory retains important information for the user between conversations and retrieves details necessary for tasks. Active Focus is a feature that allows the user to continue toward the goal even when the task becomes difficult. If an error occurs, it attempts alternative methods and requests an explanation from the user if necessary. It also performs confirmation before executing important actions.<\/p>\n\n\n\n
The central elements that constitute the user experience are memory and task continuity. It is designed to reduce the burden of explaining background every time and to break down long requests into multiple steps for processing. The developer page also introduces features for manipulating browsers, desktops, and mobile apps. Use cases presented by Meta include processing insurance claims, understanding lease agreements, and app development. The research blog showcases an example of creating wedding seating charts tailored to complex conditions.<\/p>\n\n\n\n
Muse Wonder is available through the Meta Model API and the coding tool Muse Code. The Muse Wonder pool is provided to Muse Code subscribers. The contextual processing scope of both models is 1 million tokens. Meta plans to apply Wonder to its personal agent service, Muse, in the future and provide it to subscribers first. It stated that the system has learned privacy protection and responses to malicious instructions tailored to user environments involving messages, schedules, and personal documents. An additional safety assessment report is scheduled to be released at a later date.<\/p>\n\n\n\n