AI News · 4 min read
Muse and Meta AI get Muse Wonder's Active Memory for everyone
Meta moved Muse and Meta AI to Muse Wonder-trained models on Nov 6, 2026, adding Active Memory and Active Focus for all users. Meta's evals and our take.
What Meta shipped on November 6, 2026
Meta published Bringing Specialized Models to Muse and Meta AI ↗ on November 6, 2026. Muse, Meta's personal agent app, and the Meta AI assistant now run on "specialized versions of our existing models, trained using Muse Wonder." Wonder is the frontier model family Meta introduced on October 12 ↗. The New Stack ↗ confirmed that the new models go to all Muse and Meta AI users, while Wonder itself stays with subscribers for harder tasks.
This is the third Muse update from Meta in a week. It follows Muse Realtime Voice and avatars and Muse Spaces. Those two added new screens to the app. This one swaps the model underneath, so it touches every conversation.
Two Wonder features come along:
Active Memory. Meta says the models are trained "to recognize what type of information is temporary versus permanent, to reason across time, and to distinguish between what is still relevant and what has become stale."
Active Focus. Meta says the models were trained on "a variety of bottlenecks reflective of the real world, so that they can try harder, recover, or find alternatives on your behalf."
Meta's own numbers (internal evals, not independently verified):
| Metric (Meta) | Previous Muse model | New model |
|---|---|---|
| Personal Superintelligence Index | 38% | 47% |
| Active Memory eval: tasks correctly using critical context | 76% | 90% |
| Forgetting critical info (Muse) | baseline | 2.4x reduction |
| Finishing everyday tasks when issues arise (Muse) | baseline | 67% improvement |
Meta also says the models are better at deciding when to work alone and when to check in, and at taking "steers and clarifications in the middle of running tasks without getting confused." In Meta AI it pitches the same gains as a better thinking partner that "can break complex problems into clear steps" and keep up "when you change direction mid-conversation."
What Meta didn't say: the new models' names, sizes, or how it trained them from Wonder. The New Stack notes that Meta "hasn't detailed its method," and that the 67% figure came without a baseline.
What people are saying on X
Meta's official post (@AIatMeta ↗):
"Today we're bringing specialized models trained using Muse Wonder to @Muse and Meta AI, giving users access to some of Muse Wonder's frontier capabilities for complex tasks at work and in everyday life."
Meta's Alexandr Wang (@alexandr_wang ↗):
"starting today, muse and meta ai will run on new specialized models trained using muse wonder.
ACTIVE MEMORY: models can reason across time and better remember what matters
ACTIVE FOCUS: enabling muse to continue working towards a goal even when met with roadblocks"
TestingCatalog's summary drew the most engagement among third-party posts we checked (@testingcatalog ↗):
"Muse and Meta AI apps now run on new specialized models trained with Muse Wonder. Both Active Memory and Active Focus, introduced in Muse Wonder, are now available in Muse and Meta AI."
A fair caveat (@de2pressed ↗):
"Active Memory hits 90% vs 76% on critical context; Personal Superintelligence Index rises to 47% from 38%. Internal, self-reported metrics."
Our take: who should care
Free users get the most out of this. Memory that carries across sessions and an agent that keeps going when a step fails have mostly been paid-tier selling points. The New Stack notes that Muse has a free tier, while OpenAI's Dots and Anthropic's Duet sit on those vendors' top plans. Meta is now putting a Wonder-trained model in front of everyone on that free tier, plus Meta AI inside WhatsApp, Instagram, Facebook and Messenger. If you tried Muse earlier and it lost track of what you told it last week, that's the exact problem Meta claims to have fixed. It's worth a second try.
Power users: this isn't Wonder. Subscribers keep the full model for harder work, and developers get Wonder through the Meta Model API. The New Stack lists it at $4 per million input tokens and $20 per million output. If your work is coding or long agent runs, the consumer upgrade won't replace that.
Skepticism is warranted on the numbers. The Personal Superintelligence Index and the Active Memory eval are Meta's own benchmarks with no outside replication. A 9-point index gain and a 14-point memory gain sound good, but you can't compare them with ChatGPT or Claude memory on equal terms yet.
Privacy check: better memory means the assistant keeps more about you between chats. Look at what Muse and Meta AI store before you rely on it for anything sensitive. Meta's post doesn't add new memory controls.
Comparing assistants? Start with ChatGPT vs Claude vs Gemini and our best ChatGPT alternatives. For Meta's open-weight history, see our Llama 4 review.
Sources
Meta AI Research: Bringing Specialized Models to Muse and Meta AI ↗, November 6, 2026
Meta: Introducing Muse Wonder and Muse Wonder Full ↗, October 12, 2026
The New Stack: Meta's best AI model now trains the one most of its users get ↗, November 6, 2026