Meta rolled out its Muse Wonder AI models on Monday for users to handle tasks like filing insurance claims, reviewing leases, and building software.
In a new blog post, Meta called its Muse Wonder and Muse Wonder Full models its "most capable" to date. CEO Mark Zuckerberg referred to them as a "major leap" toward personal superintelligence.
The biggest upgrades for the new Muse agents are what Meta is calling "Active Memory" and "Active Focus."
Active Memory allows the models to carry "important context about you across conversations" and retrieve details on particular tasks. In a launch video posted by Chief AI Officer Alexander Wang, Meta said that means the agent can remember a user's habits, preferences, and past conversations spanning months of inputs.
Active Focus, meanwhile, is how the agents work through roadblocks. Meta says the models can try different approaches when something goes wrong and can "adapt to new information" while clarifying things with the user along the way.
In Meta's self-published benchmark tests, its Muse models beat out Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Astra on memory tests and "long-running software projects." However, it trailed those rivals on other tests involving health questions and coding tasks.
Developers can try out Muse Wonder through Meta's API and Muse Code, while the Wonder Full model is only available to Muse Code subscribers. Meta also says it plans to add Wonder to its Muse assistant.