Meta Launches Muse Wonder Models with Active Memory and Focus Features
Meta introduced the Muse Wonder and Muse Wonder Full AI models, featuring new Active Memory and Active Focus capabilities designed to enhance agent performance on long-horizon tasks, claiming state-of-the-art results.
BedeutungStrukturellBeweisE3 überprüfbarAufbereitungSchnell
Meta officially launched the Muse Wonder series on October 12, 2026, comprising Muse Wonder and Muse Wonder Full. The company states these models achieve state-of-the-art performance on personal and professional agentic tasks, marking a significant milestone in its ground-up AI overhaul initiated fourteen months ago.
The release highlights two key architectural features: Active Memory and Active Focus. Active Memory enables the model to retain critical context across conversations and actively retrieve relevant details. Active Focus allows the system to sustain effort on challenging goals by trying alternative approaches, working through blockers, and adapting to new information without constant human intervention.
Muse Wonder is now available via the Meta Model API and Muse Code, while Muse Wonder Full is currently restricted to Muse Code subscribers. Meta reports training occurred at its gigawatt-scale Prometheus data center. While the company cites benchmark results supporting its frontier claims, these figures remain self-reported and await independent reproduction or detailed third-party validation.