Meta's Muse Wonder brings active memory and focus to subscribers and API users, matching frontier AI performance
Meta has launched Muse Wonder and Muse Wonder Full, introducing active memory and focus capabilities to its personal agent platform for subscribers and API users.

Key points
- Meta released Muse Wonder and Muse Wonder Full, claiming they match frontier performance in coding and complex tasks.
- The new models feature active memory for cross-conversation context and active focus for persistent goal pursuit.
- Muse Wonder is available via API and Muse Code, while Muse Wonder Full requires higher-tier subscriptions.
Meta has officially released two new artificial intelligence models, Muse Wonder and Muse Wonder Full, to its user base. According to a report by Engadget, the company states these models represent a significant advancement toward personal superintelligence, marking the culmination of more than a year of intensive development and tens of billions of dollars in investment. The launch targets both paid subscribers and developers accessing the platform through its application programming interface.
The primary technical distinction for these models is the introduction of active memory and active focus capabilities. Meta describes active memory as the ability to retain relevant details across multiple conversations, ensuring that important context is available when needed for specific tasks. Active focus allows the agent to persist in working toward a user's objective even when the task becomes difficult or complex, rather than abandoning the effort.
Technical Capabilities And Performance Claims
Meta asserts that Muse Wonder and Muse Wonder Full have reached parity with the leading models in the industry. The company highlights that the systems are particularly effective at handling legal, financial, and health-related workflows. Additionally, Meta claims the models deliver frontier-level performance in coding tasks, a critical metric for developer adoption and enterprise utility. These capabilities are intended to help users accomplish more over extended periods by maintaining context and perseverance.
In an interview with the Financial Times, Meta CEO Mark Zuckerberg explained that reaching this milestone required substantial computational resources and rigorous scientific work from the company's superintelligence team. The company expects to spend up to $145 billion this year to support its ambitious artificial intelligence initiatives, underscoring the scale of the infrastructure and research efforts behind these model releases.
Availability And Pricing Tiers
The rollout is beginning with paid subscribers, with broader access available through Meta's API and Muse Code platforms. Muse Wonder is priced at $4 per million input tokens and $20 per million output tokens for API users. This model is also accessible within the Muse Code environment for standard users.
Muse Wonder Full, the more advanced variant, is restricted to users on higher subscription tiers. It is available only to those subscribed to the $50 per month power usage plan or the $200 per month maximum usage plan of Muse Code. This tiered approach ensures that the most resource-intensive model is reserved for users willing to pay for increased usage limits and capabilities.
What should you do?
Security and technology professionals should monitor the performance and safety outputs of these new models, particularly given their increased autonomy and persistence. Evaluate whether the active memory features introduce new data retention risks for sensitive legal or financial information. Review API usage policies to ensure compliance with data privacy regulations when integrating these models into enterprise workflows.
What to do and how to stay safe: Meta
- Assess how active memory features handle sensitive data retention across long-term user sessions.
- Monitor API usage costs and token limits to prevent unexpected billing spikes for enterprise deployments.
- Test the models' persistence in coding tasks to verify accuracy and identify potential hallucination risks.
- Review subscription tier requirements to ensure access aligns with organizational budget and usage needs.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What are the main new features of Meta's Muse Wonder models?
The models introduce active memory for retaining context across conversations and active focus for persisting through difficult tasks.
How much does Meta plan to spend on AI this year?
Meta expects to spend as much as $145 billion this year to fuel its artificial intelligence plans.
Who can access Muse Wonder Full?
Muse Wonder Full is available to subscribers of the $50/month power usage or $200/month maximum usage tiers of Muse Code.


