Meta Launches Muse Wonder and Muse Wonder Full: AI Models Built to Complete Real Tasks for Users

What Meta announced

According to Meta's official press release, on October 12, 2026 Meta announced Muse Wonder and Muse Wonder Full, calling them its most capable models to date. Meta says the models build on the foundation laid by Muse Spark, and that it began overhauling its AI work 14 months ago.

Meta says the models deliver state-of-the-art performance on personal and professional "agentic tasks". An agentic task is one where AI does more than answer questions: it acts like an agent and completes work for the user step by step, such as handling an insurance claim, understanding a lease, or building a working app. This is Meta's own description and has not been verified by third parties.

Availability and key features

The API (application programming interface) in the table is the channel that lets developers plug the model into their own software, so for now it is mainly developers who can use the models directly.

Meta says the new models have Active Memory, which remembers what matters to the user across conversations and brings details back up in relevant tasks. Meta also says they have Active Focus: when a task gets hard, the model keeps working toward the goal, tries other approaches and recovers from mistakes, asks for clarification when needed, and confirms before doing anything important.

Meta says the models bring these capabilities to professional domains such as law, finance and health, along with frontier coding performance.

Safety notes and what they mean for general readers

Meta says personal agents may need to work across messages, calendars and documents, so the Muse Wonder family was trained to avoid sharing private information without permission, to refuse to help with actually harmful actions, and to improve security and strengthen defenses against cyberattacks. Meta also says developers and security teams can use such models to find flaws in their own code, and that Meta ensured the models can still do this kind of work.

Meta further says its most powerful AI models are trained, evaluated and released according to the Meta Superintelligence Scaling Framework, and that a full report with more safety evaluations will be available soon. Until that report and independent evaluations appear, readers should treat Meta's capability and safety claims as the vendor's own statements.

For everyday users, the more immediate effect in the short term is that Meta says Muse Wonder will roll out in Muse, subscribers first. If these agent features later reach messages, calendars or documents, check permissions and privacy settings before using them.

FAQ

Browse the latest news in this topic

• Tech news