Meta Platforms introduced Muse Wonder and Muse Wonder Full on Monday, presenting them as its most capable artificial-intelligence models and a step toward the company’s goal of building personal AI systems that can handle complex tasks. Meta Chief Executive Mark Zuckerberg has made that broader ambition a priority, and Meta shares climbed after the announcement.
The models are designed for personal and professional “agentic” work, including navigating insurance claims, interpreting leases, handling financial or health information and building software. Meta says Muse Wonder can retain relevant information across conversations through Active Memory, while Active Focus is intended to keep working toward a goal, try alternate approaches, recover from errors and ask for clarification before important actions.
Muse Wonder is available to developers through Meta’s Model API and to users of Muse Code. Muse Wonder Full, the more advanced version, is available to Muse Code subscribers. Meta said Muse Wonder will soon power features in its Muse assistant, initially for subscribers, and may later support specialized models and other products.
Meta said both new models were trained at its gigawatt-scale Prometheus data center and produced state-of-the-art results across evaluations involving agents, memory and coding. The company also said it trained the models to protect private information, resist harmful requests and support defensive cybersecurity work; additional safety evaluations are due in a fuller report. Independent testing will determine how the models compare with competing systems, while Meta’s next challenge is extending the technology from developers and subscribers to its broader user base.