Meta Rolls Out New Muse Models With Better Memory to All Users

On 6 November 2026, Meta began shipping new models to all users of its Muse assistant and Meta AI. According to Meta, the models bring two features from its larger Muse Wonder model, called Active Memory and Active Focus, down to the models everyone uses.

What Meta says changed

• The new models are 2.4 times less likely to forget critical information in a conversation.
• In Meta's tests they used important context correctly in 90 percent of tasks, up from 76 percent for the previous model.
• Subscribers still get Muse Wonder itself for harder tasks.

Mark Zuckerberg announced the rollout on X. All of these figures come from Meta's own testing. No independent benchmark has confirmed them, so they show direction, not a guarantee.

Why forgetting is the problem to solve

Anyone who has used an AI assistant for a long task knows the failure: you give it a rule at the start, and twenty messages later it ignores the rule. Better memory and focus matter most for exactly the jobs people increasingly hand to assistants, such as editing a long document, planning a trip with many constraints or following a style guide.

What to watch

Check it on your own work. A good test is to give the assistant three firm rules, chat for a while, then see whether it still follows them. If you create content with Meta AI, keep your important instructions in one saved message you can paste back in, whatever the model claims.

A simple three-step memory test you can run

1. Set rules early. Start a chat with three clear instructions, for example: answer in short sentences, always give a Telugu and an English title, and never use emojis.
2. Add distance. Ask ten unrelated questions, so the rules sit far back in the conversation.
3. Check the result. Ask for a new task and count how many of the three rules the assistant still follows.

Run the same test on the older and newer model if you can. If the newer one follows more rules, the memory claim holds up for your kind of work, and that is worth more than any company benchmark.

What this means for creators

People who write prompts for a living, such as the ones in our prompt library, rely on assistants following long, detailed instructions. Better context handling should reduce the need to repeat style rules on every request, but it will not replace writing clear prompts. A well-structured prompt still gives the best result with any model, old or new.

Rollouts like this usually reach different countries and apps at different times, so you may not see the change in every Meta app today.