Meta brings Muse Wonder's Active Memory and Active Focus to Muse and Meta AI

Meta's @AIatMeta account announced that specialized models trained using Muse Wonder, Meta Superintelligence Labs' frontier model, are now live in the Muse app and Meta AI. The goal is to bring some of Muse Wonder's capabilities to consumer and work tasks without serving the full model to everyone.

The two named capabilities are:

- Active Memory: retains relevant context across past conversations, so the assistant builds a working understanding of the user.

- Active Focus: keeps multi-step reasoning goal-directed, aimed at the drift that affects long agentic tasks.

The approach, training smaller specialized models from a frontier "teacher," is a cost-effective way to deliver frontier-like behavior at Meta's scale of billions of users.

Memory is a competitive front this week. Anthropic published an engineering post on how "dreaming" helps Claude consolidate what it has learned, which drew about 12,400 likes. OpenAI and Google have had persistent memory for some time. Meta's advantage is distribution across WhatsApp, Instagram and Facebook.

The launch also landed in the same week that Meta had to pause its controversial AI-training terms, so a feature that retains more user context will draw scrutiny. Meta did not publish benchmarks or detail what data Active Memory stores, how long it keeps it, or how users can control it. Those are the questions to watch, especially in the EU.