<\/p>\n\n\n\n
Meta<\/a> has introduced Muse Wonder and Muse Wonder Full, its most capable models to date. The company says they deliver state-of-the-art performance on personal and professional agentic tasks. Examples include navigating an insurance claim, making sense of a lease, or building a working app.<\/p>\n\n\n\n Built on Recent AI Overhaul<\/strong><\/p>\n\n\n\n Fourteen months ago, Meta<\/a> began an overhaul of its AI efforts. The company says it has moved from rebuilding its foundations to pushing the frontier. The Muse Wonder series builds on the groundwork laid with Muse Spark<\/a>.<\/p>\n\n\n\n Designed for Real-World Needs<\/strong><\/p>\n\n\n\n Tasks like an insurance claim, a lease or personal finances share one requirement. The model must interpret the information in front of it and work through the problem.<\/p>\n\n\n\n Meta says the new models bring that ability to expert domains, including law, finance and health. They also offer frontier coding performance. That supports people building anything from a game to an app for a new business.<\/p>\n\n\n\n Active Memory and Active Focus<\/strong><\/p>\n\n\n\n Real work also unfolds over time. For that reason, Meta built the models with Active Memory. This feature carries what matters to the user across conversations. It also brings back the details relevant to the task at hand.<\/p>\n\n\n\n The models also include Active Focus. This lets them keep working toward a goal when a task gets hard. They can try other approaches, recover from errors and follow through. They also ask for clarification when needed. In addition, they check in before acting on anything important.<\/p>\n\n\n\n Safety Approach<\/strong><\/p>\n\n\n\n A personal agent can take on real tasks. This often means working across messages, calendars and documents. Therefore, Meta trained the models to avoid sharing private information without permission.<\/p>\n\n\n\n The company also trained them to turn down requests to assist with actually harmful acts. Meta says it is working to improve security and strengthen defenses against cyber attacks.<\/p>\n\n\n\n At the same time, developers and security teams can use models like Muse Wonder to find flaws in their own code. Meta says it made sure the models can still do that work.<\/p>\n\n\n\n Meta trains, evaluates and releases its most capable AI models in line with its Meta Superintelligence Scaling Framework<\/a>. A full report with additional safety evaluations will be available soon.<\/p>\n\n\n\n Availability<\/strong><\/p>\n\n\n\n Muse Wonder is available today for developers and builders. Access is through the Meta Model API and Muse Code. Muse Wonder Full is available today for Muse Code subscribers.<\/p>\n\n\n\n Muse Wonder will soon power experiences in Muse<\/a>. The rollout will start with subscribers.<\/p>\n\n\n\n Looking Ahead<\/strong><\/p>\n\n\n\n Meta describes the Muse Wonder launch as a major step toward personal superintelligence for everyone. The company adds that much more is to come.<\/p>\n\n\n\n