Meta brings Muse Wonder models to Muse and Meta AI
According to AI at Meta on X and the Meta research blog, specialized Muse Wonder models with Active Memory and Active Focus are rolling out to Muse and Meta AI, showing 2.4x less forgetting and 67% better recovery on evolving tasks.
SourceAnalysis
On November 6, 2026, AI at Meta announced the rollout of specialized models trained using Muse Wonder to both Muse and Meta AI platforms.
These models incorporate Active Memory, which recognizes when earlier information remains relevant versus stale, and Active Focus, which aids recovery from setbacks when constraints shift. According to the announcement from AI at Meta, Muse shows a 2.4x reduction in forgetting critical goal-related information and a 67% improvement in adapting when plans change. The updates also enable Meta AI to decompose complex problems into steps, pose sharper follow-up questions, and adjust direction more effectively.
AI at Meta
@AIatMetaTogether with the AI community, we are pushing the boundaries of what’s possible through open science to create a more connected world.