Meta's Muse now forgets 2.4x less with Active Memory

Short answer

Since November 6, 2026, Meta's Muse agent and Meta AI run on new models trained with Muse Wonder, Meta's most capable model. They bring two features: Active Memory, which helps the assistant remember what matters from past chats, and Active Focus, which helps it keep going when a task hits a snag. Meta says that in Muse the new models forget critical information 2.4x less often. Everyone using Muse and Meta AI gets them; Meta's post does not list countries.

What happened

In a post on its research blog, Meta said it built "specialized versions" of its existing models, trained using Muse Wonder, and is rolling them out to people who use Muse and Meta AI. Muse Wonder itself launched in October; according to The New Stack, subscribers still get Wonder for harder tasks.

According to Meta (also reported by The New Stack and SiliconReport):

• Active Memory: the models are trained to tell temporary information from permanent, to "reason across time", and to notice when an older detail has become stale.
• Memory score: on Meta's internal Active Memory test, Muse now uses critical context correctly in 90% of tasks, up from 76%.
• Active Focus: Meta reports a 67% improvement in Muse finishing everyday tasks when something unexpected goes wrong. The New Stack notes Meta did not share the baseline for that figure.
• In Meta AI: Meta says the update makes it better at breaking problems into steps, asking sharper follow-up questions and keeping up when you change direction mid-conversation.

All numbers are Meta's own evaluations, not independent tests.

Why people are talking about it

The news spread quickly on X, with posts from Mark Zuckerberg, Alexandr Wang and Meta's AI team, and it made Techmeme's front page. Forgetting is one of the most common complaints about AI assistants, so a claim of "less repeating yourself" lands with everyday users, not only with developers. It also matters for scale: Meta AI runs inside WhatsApp, Instagram, Facebook and Messenger.

What it means if you use AI

• Fewer repeats. If the claims hold up, you should need to restate your goals, preferences and plans less often.
• Memory is a choice, too. An assistant that remembers more is more useful, so it is worth knowing what it keeps. Check whether your app lets you review or delete what it saves.
• Better follow-through. Active Focus is aimed at tasks that break halfway, like a booking where a website fails.
• It is still AI. Better memory makes conversations feel smoother, but it is software that follows your settings, not a person.

Quick checklist

1. Update the Muse or Meta AI app so you are on the latest version.
2. Mention a preference in one chat, then test whether it shows up later.
3. Look in the app's settings for memory or chat-history controls.
4. Change your mind mid-task and see whether it keeps up.
5. Double-check anything important it remembers for you, like dates or numbers.

Sources

• Meta AI Research (Nov 6, 2026): Bringing Specialized Models to Muse and Meta AI (primary)
• Meta AI Research: Muse Wonder, a frontier model for personal and professional AI agents
• The New Stack (Nov 6, 2026): Meta's best AI model now trains the one most of its users get
• SiliconReport (Nov 6, 2026): Meta rolls out specialized AI models trained using Muse Wonder

Sources read on November 7, 2026 (Europe/Prague).

Related reading:

• Meta Muse gets real-time voice and expressive avatars
• Study: 96% of ChatGPT memories were saved unasked
• Why does AI forget what you told it?

Cover photo: rawpixel, "Texting over iced coffee", via Wikimedia Commons (Unsplash), [CC0](https://creativecommons.org/publicdomain/zero/1.0/).