Meta launches Muse Wonder models to power agents with memory and recovery
NO IMAGE
AI Agents & Automation
Meta launches Muse Wonder models to power agents with memory and recovery
Meta NewsroomOct 12, 2026
Meta introduced Muse Wonder models with persistent memory and task-recovery features, making them available to developers and some subscribers.
Meta introduced Muse Wonder and Muse Wonder Full on October 12, describing them as its most capable models to date for personal and professional agentic tasks. The company says the models are designed for multi-step work such as understanding a lease, navigating an insurance claim, handling financial information and building software. Their release marks a prominent addition to Meta’s effort to compete in advanced AI after rebuilding its model-development program.
The models’ key advertised features are Active Memory, which carries relevant information across conversations, and Active Focus, which helps a model continue working when a task becomes difficult, try alternative approaches and recover from errors. Meta says the models can ask for clarification and check with users before taking important actions. These features target tasks that extend beyond a single question-and-answer exchange and require maintaining context over time.
Availability is split by product and audience. Muse Wonder is available to developers through Meta’s Model API and Muse Code, while Muse Wonder Full is available to Muse Code subscribers. Meta says Muse Wonder will later power more experiences in its Muse assistant, initially rolling out to subscribers.
The release also raises the stakes for safeguards as models gain access to messages, calendars, documents and other personal or business context. Meta says it trained the models to avoid sharing private information without permission and to refuse harmful requests. It also says developers and security teams can use them to find vulnerabilities in their own code, while noting that the full report of additional safety evaluations would be published later. The company’s stated safeguards and performance claims are its own; independent evaluation will help establish how reliably the models perform in real deployments.
Industry Impact
Agents built for longer tasks
Software development, business workflows, personal assistance
Overview
Meta’s Muse Wonder models add persistent memory and task-recovery features to its agent offering, targeting work that spans multiple steps and conversations. Developers can access Muse Wonder through Meta’s Model API and Muse Code, while Muse Wonder Full is limited to Muse Code subscribers. For software developers and product teams, this creates a new option to test agents on tasks involving documents, claims, financial information, or code. Adoption will depend on performance and safeguards in real deployments, which have not yet been independently established.
Key Impact Points
Persistent memory is intended to let agents carry relevant context across conversations.
Active Focus is designed to help models try alternatives and recover when tasks become difficult.
Access is tiered across developers and subscribers, while Meta says broader assistant rollout will begin with subscribers.
Opportunities
Build and test persistent-context agents
Software development, insurance, financial workflows, business operations
The release gives developers a basis to explore agents that maintain context across interactions and attempt to recover from errors. Potential applications supported by the article include assisting with lease understanding, insurance claims, financial information, and software development. Teams could prototype workflows through the Model API or Muse Code, then assess whether clarification prompts, user approval before important actions, and privacy safeguards are adequate for their use case. Meta has not disclosed pricing, commercial terms, or independent performance results, so any business case should remain conditional on evaluation, deployment requirements, and user acceptance.
Multi-step workflow agents
Problem: Tasks such as understanding a lease or navigating an insurance claim can require context and follow-through beyond a single question-and-answer exchange.
Solution: Developers can evaluate Muse Wonder for agents that retain relevant information across conversations and attempt alternative approaches when work becomes difficult.
Target Users: Developers and teams building personal or business workflow assistants.
Business Value: Could support more continuous assistance on multi-step tasks, subject to reliable performance and suitable safeguards.
Revenue Potential: Potential depends on adoption and deployment economics; the article provides no pricing or revenue details.
