Hopp til hovedinnhold

Meta rolls Active Memory and Active Focus into Muse and Meta AI

Meta is rolling Active Memory and Active Focus into Muse and Meta AI. It reports 2.4× less forgetting and 67% better task completion — company figures, not independently verified.

Av Joachim HøgbyKilde: Meta AI Research

Meta is rolling out specialized models with Active Memory and Active Focus to Muse and Meta AI. The company reports 2.4× less forgetting of critical context in Muse and a 67% improvement in completing everyday tasks when unexpected obstacles appear. Those figures are Meta’s own measurements, not independently verified.

The announcement came via Meta’s AI research blog and the official @AIatMeta account on Friday 6 November 2026 (18:14 UTC on X). This is not a launch of Muse Wonder itself. Meta describes specialized versions of existing models, trained using Muse Wonder, now rolling out to people who use the Muse app and Meta AI.

What is actually shipping

According to Meta, Active Memory and Active Focus were first introduced with Muse Wonder. Both capabilities are now moving into the models that power Muse and Meta AI:

  • Active Memory is meant to recognize when earlier information becomes relevant, distinguish temporary from permanent details, and separate what still matters from what has gone stale. Meta says improved Muse models show a 2.4× reduction in forgetting information critical to user goals, and that tasks correctly using critical context rose from 76% to 90% on Meta’s Active Memory evaluation.
  • Active Focus is meant to help Muse keep working toward a goal when unexpected problems appear — try harder, recover, or find alternatives. Meta reports a 67% improvement in completing everyday tasks with such setbacks, without publishing the underlying completion rates.

Meta also says Muse’s Personal Superintelligence Index rose from 38% to 47% in its own measurements. For Meta AI, the company describes clearer steps, sharper follow-up questions, and better handling when the user changes direction mid-conversation — without separate numbers for Meta AI.

What Meta does not claim

Meta does not say that Muse Wonder Full or the full Wonder series is now the default for everyone in Muse. The post covers specialized versions of existing models. The results are the company’s own measurements; the blog does not present an independent evaluation. Availability outside the United States, pricing, and any subscription requirements are not specified in this announcement.

Why it matters for leaders

  • Agent memory becomes a product requirement, not just a demo. When Meta ties measurable “memory” and “focus” to products leaders already evaluate (Muse/Meta AI), pressure rises on vendors to document context stability — not only leaderboard peaks.
  • The numbers are vendor claims. 2.4× less forgetting and 67% better completion are Meta’s figures. Ask for method, sample and independent review before using them in procurement or board materials.
  • Long-context governance. Better memory across conversations also means more stored context about employees, customers and decisions. Map where data lives, retention, and who can delete or export before a broad rollout.
  • Competition for the “personal agent”. Meta is positioning Muse against other personal/agentic assistants. For organisations already testing Muse or Meta AI, this is a concrete capability step — but without clear enterprise controls in this post, keep any pilot bounded.
  • Separate Wonder from product rollout. Wonder launched earlier. Today’s story is transferring selected Wonder capabilities into broader product surfaces. Do not conflate them in risk assessment.

Sources and media

Illustration: AI-generated with Grok Bot / hogby.ai

Mer om AI

Oppskriften. Strategien. Hver dag.

AI-nyheter for ledere. Kuratert av en CIO som bygger det selv. Daglig i innboksen.