Meta launched two new AI models: Muse Wonder and Muse Wonder Full. The company calls them its most capable models so far.
As expected, these models are built for agentic tasks. This means they can handle multi-step jobs, adjust when problems come up, and keep working toward a goal over time. They were trained at Meta’s Prometheus data center.
Two new features power them. Active Memory lets the models recall important details across conversations. Active Focus helps them push through obstacles, recover from errors, and adapt to new information. They can also ask users for clarification before taking major actions.
The models handle specialized work in law, finance, and health. For coding, they can track requirements, edit multiple parts of a codebase, and test their own work.
On safety, Meta said the models guard private information, refuse harmful requests, and resist adversarial instructions. The company claims its strongest safeguards yet against chemical, biological, radiological, and nuclear misuse, plus cyber misuse.
Muse Wonder is available to developers through the Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers. Meta also plans to bring Muse Wonder to its Muse app for subscribers.