Meta has launched the new Muse Wonder and Muse Wonder Full premium AI models on Monday. The models build upon the Muse Spark AI model, launched earlier this year. The tech giant says the new models, designed for personal and professional agentic tasks, offer specialised knowledge to solve problems for users at work and in daily life. Muse Wonder combines Active Memory and Active Focus to help agents recognise when earlier information becomes relevant and work through problems.
Meta Announces Muse Wonder, Wonder Full
In a blog post, Meta announced the launch of the Muse Wonder series with Muse Wonder and Muse Wonder Full. The new AI models are claimed to be the company’s “most capable models that deliver state-of-the-art performance on personal and professional agentic tasks”. They bring together Active Memory and Active Focus. The company says it can tackle everything from understanding a lease to making sense of financial or health information.
Muse Wonder is available now through Meta’s Model API and Muse Code. The Wonder Full is available to Muse Code subscribers. The company confirmed that Muse Wonder will soon power experiences in Muse and will roll out to subscribers first.
Meta has highlighted Active Memory and Active Focus features of the Muse Wonder series. The former carries important context about the user across conversations and retrieves relevant details to complete a task. Active Focus, on the other hand, tries alternative approaches, working through blockers, and adapting to new information.
The models will ask for clarification when needed and check in before taking important actions. Meta says it evaluates focus through FocusBench and execution through SWE-Marathon.
Muse Wonder and Wonder Full are also designed to deliver frontier coding performance, supporting projects from small applications to complex software.
Meta has also shared benchmark and testing scores for Wonder and Wonder Full with other available models. As per the Harvey LAB-AA legal agent benchmark, Muse Wonder Full scored 13.1, outperforming baseline Muse Wonder (10.0) and GPT-6 Astra (8.6). For long-horizon coding tasks evaluated on DeepSWE 1.1, the Muse Wonder Full model scored 79.0, with Muse Wonder getting 78.8.
Meta claims that Muse Wonder offers our strongest safeguards to date against CBRN (chemical, biological, radiological, and nuclear) and cyber misuse, while preserving access to frontier cyber capabilities for defensive use, per our Meta Superintelligence Scaling Framework.
