2026-10-13 — views
Meta Unveils Muse Wonder and Muse Wonder Full for Agentic AI Tasks
Read this because Meta's new Muse Wonder models offer state-of-the-art performance for professional and personal agentic tasks, featuring active memory and safety protocols.
Meta introduces Muse Wonder and Muse Wonder Full, frontier models designed for complex agentic tasks with active memory and focus capabilities.
Meta Introduces Frontier Models for Agents
Meta has officially introduced Muse Wonder and Muse Wonder Full, describing them as the company’s most capable models to date. These new additions are specifically built to deliver state-of-the-art performance on personal and professional agentic tasks. The models are designed to handle complex real-world needs, such as navigating insurance claims, interpreting lease agreements, and analyzing personal finances. Additionally, they provide frontier coding performance, enabling developers to build everything from games to working applications for new businesses.
Capabilities: Active Memory and Active Focus
A key differentiator for the Muse Wonder series is its ability to handle tasks that unfold over time. The models feature Active Memory, which allows them to carry relevant information across conversations and retrieve details pertinent to the current task. This is complemented by Active Focus, a capability that enables the model to persist toward a goal even when a task becomes difficult. The models can attempt alternative approaches, recover from errors, ask for clarification when necessary, and check in before executing important actions. These models build upon the groundwork laid by the previously released Muse Spark.
Safety and Security Standards
Meta emphasizes that personal agents often work across sensitive user data, including messages, calendars, and documents. To address this, the Muse Wonder family was trained to avoid sharing private information without permission and to decline requests to assist in harmful acts. The models also work to improve security and strengthen defenses against cyber attacks. However, Meta notes that developers and security teams can still use these models to identify flaws in their own code and strengthen systems. The training, evaluation, and release of these models adhere to the standards set out in the Meta Superintelligence Scaling Framework. A full report with additional safety evaluations is scheduled to be available soon.
Availability and Access
Muse Wonder is available today for developers and builders through the Meta Model API and Muse Code. Muse Wonder Full is currently available for Muse Code subscribers. The models will soon power more experiences within the Muse platform, with an initial rollout to subscribers. This release marks a significant step in Meta’s efforts, which began fourteen months ago with an overhaul of their AI foundations, moving from rebuilding infrastructure to pushing the frontier of AI capability.
Practitioner note
Developers should integrate Muse Wonder via the Meta Model API to test agentic workflows requiring long-context memory. Evaluate the Active Focus capabilities to ensure robust error recovery in complex, multi-step automation tasks.