Fourteen months after overhauling its AI efforts, Meta is introducing the Muse Wonder series, which includes Muse Wonder and Muse Wonder Full. The company calls them its most capable models to date. Built on the groundwork laid with Muse Spark, they deliver what Meta describes as state-of-the-art performance on personal and professional agentic tasks. Meta says the overhaul moved its work from rebuilding foundations to pushing the frontier.
Built for Real-World Work
Meta points to everyday jobs such as navigating an insurance claim, understanding a lease, or making sense of someone's finances. In each case, a model has to interpret the information in front of it and work through the problem. Muse Wonder and Muse Wonder Full are aimed at expert domains including law, finance, and health. They also bring frontier coding performance to people building everything from a game to an app for a new business.
Real work also unfolds over time, so Meta built the models with two capabilities. Active Memory carries what matters to a user across conversations and brings back the details relevant to the task at hand. Active Focus keeps the model working toward a goal when a task gets hard. It can try other approaches, recover from errors, and follow through. It also asks for clarification when needed and checks in before following through on anything important.
Safety for a Personal Agent
A personal agent can take on real tasks across a user's messages, calendars, and documents. Meta says it trained the Muse Wonder family to avoid sharing private information without permission. The models are also trained to turn down requests to assist in performing actually harmful acts, and to improve security and strengthen defenses against cyber attacks. Developers and security teams can still use models like Muse Wonder to find flaws in their own code, and Meta says it made sure the models can do that work. The company trains, evaluates, and releases its most capable models in line with its Meta Superintelligence Scaling Framework. A full report with additional safety evaluations will be available soon.
Key Features at a Glance
- Active Memory: carries relevant details across conversations, so context from earlier sessions can return when a task needs it.
- Active Focus: keeps pursuing a goal when a task gets hard, trying other approaches, recovering from errors, and following through.
- Clarification check-ins: the model asks for clarification when needed and checks in before acting on anything important.
- Expert domains: targets law, finance, and health, the kinds of work where interpreting documents matters.
- Frontier coding performance: supports people building a game or an app for a new business.
- Privacy safeguards: avoids sharing private information without permission.
- Defensive security use: developers and security teams can still find flaws in their own code.
Availability
Muse Wonder is available through the Meta Model API and Muse Code. It will soon power more of Muse, rolling out first to subscribers. Meta calls this a major step toward personal superintelligence for everyone, with more to come. The company says it will publish more on how the models were built, tested, and what they can do on its research blog.
