What happened
Meta introduced the Muse Wonder series, consisting of Muse Wonder and Muse Wonder Full. The company says the models are designed for personal and professional agentic tasks, including memory, sustained work, coding, browser use and cross-application workflows. Muse Wonder is available through the Meta Model API and Muse Code, while Muse Wonder Full is available to Muse Code subscribers. Meta says Muse Wonder will soon begin powering Muse experiences, initially for subscribers.
Meta says Muse Wonder and Muse Wonder Full are its most capable models and represent the next step in its scaling program. The company describes Active Memory as retrieving relevant personal context across conversations and Active Focus as continuing work through alternative approaches, blockers and new information while seeking clarification before important actions.
According to the source, Muse Wonder is available through the Meta Model API and Muse Code, and Muse Wonder Full is available to Muse Code subscribers. Meta says Muse Wonder will soon power Muse experiences, beginning with subscribers, as well as artifact generation and other processes requiring more intelligence. The announcement does not state API pricing, subscription prices, geographic availability or a general-release date for Muse.
Meta reports strong results across evaluations for agents, memory and coding, including legal, financial and health-related tasks, FocusBench and SWE-Marathon. It also presents demonstrations involving a playable 3D kitchen game and a browser-based running-plan workflow. These are descriptions and demonstrations supplied by Meta, not independent test results.
Source details: research.meta.ai ↗
Why it matters
The announcement places memory, sustained task execution and tool use at the center of Meta’s next AI product push. If Meta’s claims hold up, these capabilities could make its agents more useful for work involving private documents, calendars, software projects and other multi-step tasks. The source also ties the launch to safeguards for private information and cyber and CBRN misuse. However, the announcement provides no independent validation of its claims, and the full safety report is not yet available.
The release is consequential because it combines a model launch with a stated plan to embed more capable agents into a consumer product. Agents that can remember user context, operate across apps and sustain work over time could affect how people manage research, planning, coding and administrative tasks.
Those capabilities also increase the stakes around privacy, authorization and reliability. Meta says the models were trained to resist adversarial instructions, handle private information carefully and provide stronger safeguards against CBRN and cyber misuse while retaining defensive cyber capabilities. The source does not provide enough methodology or results to assess those claims independently.
Meta frames Muse Wonder as a step toward “personal superintelligence,” but that is the company’s characterization rather than an independently established fact. The practical significance will depend on real-world error rates, user controls, data practices and access conditions.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
What to watch next
Watch for the promised rollout inside Muse, publication of the additional safety evaluations and evidence about how the models perform on unfamiliar, long-running tasks. Users should also watch for clearer information about subscription pricing, geographic availability, API terms, data handling and which features are restricted to subscribers. Meta’s claims about frontier performance and alignment remain company-reported until independently reproduced.
Meta says a full report with additional safety evaluations will be available soon. That report should clarify the testing methods, threat models, failure rates and trade-offs behind the company’s safety claims.
The rollout inside Muse is also important. The source says subscribers will receive it first but does not specify timing, regions, pricing or the precise capabilities that will be enabled. API and Muse Code users may have different limits or terms.
Independent evaluations should test whether the reported performance transfers to unfamiliar tasks, long-running projects and workflows involving sensitive information. It is also worth watching whether the models ask for confirmation before consequential actions and how users can inspect or correct remembered context.