Meta Says Its New Muse Models Forget 2.4× Less
New models trained with Muse Wonder are rolling out to everyone using Muse.
Image: Meta
Meta rolled out new models for Muse on November 6, trained using Muse Wonder, its most capable model. They're going to everyone who uses Muse and Meta AI — no upgrade, no extra cost.
The headline number is memory. Meta says the new models show a 2.4× reduction in forgetting critical information, and a 67% improvement in getting unstuck when a task hits a snag. Those are Meta's own figures from its own tests.
On Meta's memory test, the new model used the right key details in 90% of tasks, up from 76%.
Forgetting is the quiet problem with AI agents. A job starts strong, then loses the thread halfway through and you're back to repeating yourself.
Models that hold the thread longer can take bigger jobs with fewer check-ins. That's the whole point of an agent.
The agents you set up today get better on their own. Same prompts, stronger engine.
Want agents doing real jobs for you? The Beginner Agent Starter Pack — 100 copy-paste setups, $4.99: go.johnveit.com/starter-pack
