Is it good at coding?

Meta reports frontier coding performance across benchmarks, with training on diverse, long-running tasks to sustain progress through repeated cycles of testing and revision on projects from small apps to complex software.

How does it handle privacy and safety?

The model family is trained to exercise discretion with private information (e.g., messages, calendars, documents) and resist misuse, including robustness to adversarial instructions and safeguards against CBRN and cyber misuse per Meta's Superintelligence Scaling Framework. A full safety evaluation report is planned.

Muse Wonder — Meta's frontier AI model for personal & professional agents

Muse Wonder is Meta's frontier AI model series (Muse Wonder and Muse Wonder Full) built for personal and professional AI agents. It combines Active Memory — carrying important context about you across conversations and retrieving relevant details — with Active Focus — sustained effort toward challenging goals, including trying alternative approaches, working through blockers, adapting to new information, asking for clarification, and checking in before important actions. The models target complex tasks at work and in everyday life, such as understanding leases, financial or health information, legal work products, long-running software projects, and personal tasks like run planning, gift design, and wedding seating charts. Meta describes the release as a step toward 'personal superintelligence.'

Key Features

Active Memory

Carries important context about you across conversations and actively retrieves the details relevant to the task at hand.

Active Focus

Sustained effort toward challenging goals: tries alternative approaches, works through blockers, adapts to new information, asks for clarification when needed, and checks in before taking important actions.

Domain expertise

Applies specialized knowledge to law, finance, and health tasks, e.g., working with source documents and following detailed instructions to produce legal work products (evaluated on Harvey's Legal Agent Benchmark).

Frontier coding

Supports projects from small applications to complex software, trained on diverse, long-running tasks to sustain progress across cycles of testing and revision (evaluated via SWE-Marathon on long-running software projects).

Breadth of generalization

Built on larger base models so post-training gains carry over to unfamiliar, out-of-distribution tasks — important for personal agents handling unanticipated combinations of goals, tools, and constraints.

Artifact generation

Will power artifact generation and processes requiring more intelligence in the Muse app.

Safety and alignment

Trained for robustness to adversarial instructions and careful handling of private information (messages, calendars, documents), with safeguards against CBRN and cyber misuse while preserving defensive cyber capabilities.

How It Works

Persistent memory and sustained execution

Muse Wonder carries context about you across conversations and retrieves relevant details on demand, then sustains effort on a goal — trying alternatives, adapting to new information, and confirming before important actions.

Works across apps and in the browser

Demonstrated working directly in a browser and across apps, e.g., building a running plan from personal preferences or designing and ordering custom swag based on photos from a personal Instagram profile.

Pros & Cons

Pros

State-of-the-art claimed performance on personal and professional agentic, memory, and coding benchmarks
Active Memory carries personal context across conversations, a key requirement for genuinely personal agents
Active Focus supports long-running tasks with checkpoints and clarification before important actions
Demonstrated breadth on out-of-distribution tasks beyond what in-distribution benchmark scores predict
Strong stated safety posture: adversarial robustness, private-information discretion, CBRN/cyber misuse safeguards

Cons

Who It's For

Best for

Developers building personal or professional AI agents via the Meta Model API
Muse Code subscribers wanting frontier coding performance on long-running software projects
Users who want an assistant that remembers context across conversations and works persistently toward goals
Workflows spanning law, finance, or health information where domain expertise matters
Everyday personal tasks like planning, shopping, and event organization across apps

Not ideal for

Users wanting a free, fully consumer-available assistant today (full Muse app integration is not yet rolled out)

Use Cases

Understanding a lease or making sense of financial or health information

Producing legal work products from source documents and detailed instructions

Building software, from small applications to complex, long-running projects

Planning a Saturday run based on personal preferences and circumstances

Designing and ordering custom swag from personal Instagram photos

Preparing a custom wedding seating chart from complex matching requirements

Building a playable 3D kitchen game from customizable image inputs

FAQ

What is Muse Wonder?

Muse Wonder is Meta's frontier AI model series for personal and professional agents, combining Active Memory (carrying context about you across conversations) with Active Focus (sustained effort toward goals). It comes in two variants: Muse Wonder and Muse Wonder Full.

What's the difference between Muse Wonder and Muse Wonder Full?

Muse Wonder Full is Meta's most capable model in the series and leads the benchmark comparisons; Muse Wonder is the standard variant. Muse Wonder Full is available to Muse Code subscribers, while Muse Wonder is available via the Meta Model API and Muse Code.

Where can I use it?

Today through the Meta Model API and Muse Code. Muse Wonder will also soon power experiences in the Muse app, rolling out to subscribers first.

What is Active Memory?

Muse Wonder