Meta has introduced two new artificial intelligence models, Muse Wonder and Muse Wonder Full, as the company expands its push into AI agents designed to handle more complex, multi-step tasks. The company announced the models on October 12, describing them as its most capable systems to date for personal and professional agentic work.
Muse Wonder is available to developers through Meta’s Model API and Muse Code, while Muse Wonder Full is available to Muse Code subscribers. Meta says the models will also begin powering features in its Muse assistant, initially for subscribers.
Designed to work across longer tasks
Meta highlights two features called Active Memory and Active Focus. Active Memory is intended to retain relevant context across conversations, while Active Focus is designed to help an agent keep working toward a goal, try alternative approaches and recover from errors.
The company says the models are aimed at practical tasks such as interpreting documents, navigating administrative processes, handling finance-related questions and building software. Those capabilities could make AI agents more useful for work that requires several connected steps rather than a single answer.
Claims will need independent testing
Meta says its new models perform strongly in areas including law, finance, health and coding. Such performance descriptions are based in part on the company’s own evaluations; broader independent testing will be important to assess how reliably the models perform in real-world use.
The launch adds to competition among major AI developers to build systems that can carry out tasks with greater continuity. It also raises familiar questions about privacy and security when agents are given access to personal documents, messages or other sensitive information. Meta says the models were trained with safeguards intended to prevent private information from being shared without permission.
