Meta Launches Muse Wonder AI Models for Complex Tasks

Meta has unveiled the next phase of its agentic AI push with Muse Wonder and Muse Wonder Full, two new models that give its Muse assistant far more processing power for demanding work. The company describes them as its most capable models yet for both personal and professional tasks.

Built for Longer, Smarter Tasks

The Muse Wonder series combines two features, Active Memory and Active Focus. Meta says these help the system track what matters to each user and stay on course toward their goals. That lets Muse handle longer, more complex projects, including some that take days to finish. Meta also reports record-setting results on key AI benchmarks. It says the models apply specialized knowledge to everyday problems, from deciphering a lease to interpreting financial or health information, and its evaluations cover law, finance and health.

For users, Muse agents are set to become considerably more capable. But that capability depends on trust. Using Muse for financial or health matters means sharing sensitive personal data with a company long criticized for its data practices.

Risks Remain Around Privacy and Human Error

Recent incidents show what is at stake. As reported by Extreme Tech, a Muse user in Toronto authorized his agent to manage a full Facebook Marketplace sale. He did not realize it would give his home address to a buyer arranging pickup. The mistake was arguably the user’s rather than the system’s. Yet because such features are so easy to switch on, similar errors could become common.

Meta says it has strengthened its security over time and is working to reassure users that their information is safe. Even so, critics argue that giving automated agents wider access to personal data, and the freedom to act on it, feels risky. They say longer testing may be needed before these tools can be considered reliably safe and valuable. Whether Muse Wonder’s improved accuracy closes that gap remains to be seen.