Meta launched Muse Wonder and Muse Wonder Full on October 12, 2026. These AI models are built for jobs that take several steps, such as working on code, going through documents or planning a project.
You give the assistant a task, and it works through the steps while keeping track of what you asked for. Meta's launch announcement describes memory, focus and coding as key parts of the new models.
Here's what those features mean in practice, along with the access and pricing details as of October 15, 2026.
What is Meta Muse Wonder?
Muse Wonder is a family of AI models for assistants that can use tools to complete a task. For example, a coding assistant might read your project files, suggest changes and run checks.
What it can actually do depends on the app you use and the access you give it. A model connected to your code editor has different tools from one running inside a chat window.
What are its main features?
Active Memory
Meta calls its memory feature Active Memory. It is meant to help an assistant remember useful details and bring them back when needed.
Imagine explaining your app's audience and design style at the start of a project. When you later ask for a new screen, remembering those details could save you from repeating the same instructions. That's a possible use for the feature; we haven't tested its memory ourselves.
Active Focus
Active Focus is about staying with a task when it takes time or runs into a problem. Meta describes a model that can try another approach, recover from errors and ask for missing information.
For a bug fix, that would mean finding the cause, making the change and checking that it works. Finishing those steps matters more than producing a long explanation of the problem.
Coding and tool use
Muse Wonder is also designed for coding tasks that involve several files. Meta describes support for using tools, working across large codebases and checking progress along the way.
A small feature in an existing project would be a sensible first trial. Give it clear requirements, then review the changes and run your tests before using the result.
Muse Wonder and Muse Wonder Full: what is different?
Muse Wonder is available through the Meta Model API and Muse Code. Its standard API model ID is muse-wonder-1.0.
Muse Wonder Full is the more capable version, according to Meta. It is available through Muse Code's Power Usage and Maximum Usage subscriptions. It is not listed as available through the Model API.
Giving the standard model more time to reason does not turn it into Wonder Full. They have separate access options.
How much does Muse Wonder cost?
The listed Standard-mode API prices for muse-wonder-1.0 are:
| Usage | Price per million tokens |
|---|---|
| Input | $4.00 |
| Cached input | $0.20 |
| Output | $20.00 |
Tokens are small pieces of text that the model reads or writes. At these rates, a request using 100,000 uncached input tokens and 10,000 output tokens would cost $0.60. Extra tools, retries and taxes can add to that amount.
These API charges are separate from Muse Code subscriptions. Check Meta's pricing documentation for current rates before choosing a plan.
How much information can it handle?
The API documentation lists a context window of 1,048,576 tokens, or roughly one million tokens. This is the amount of information the model can work with in a request.
That gives it room for long documents or substantial amounts of code. It can still miss details, so check important answers against the original material. The API documentation labels Muse Wonder as a preview.
How can you try it?
Developers can start with the Meta Model API or Muse Code. For an API project, select muse-wonder-1.0. To use Wonder Full, you need an eligible Muse Code subscription.
Before uploading private work, review the data settings. Meta says Muse Code subscriptions default to Contributor mode, which allows content to help train future models in exchange for more usage. You can turn it off. Standard-mode API access has different data terms.
Meta has also announced plans to bring the models to its Muse personal agent, with subscribers getting access first. That rollout is separate from developer access.
Is Muse Wonder useful for your work?
If you often work on code or tasks that need several rounds of instructions, start with a small project whose result you can check.
Note how much correction it needs, how long it takes and what it costs. Compare that with your usual process before moving a larger job into it.
