NewsMeta2 min read
Meta Unveils Muse Wonder, Its Most Capable Models for Agents
Muse Wonder and Muse Wonder Full are Meta’s new models for agents: with memory across conversations and checking in before acting. Where to use them.
Key points
- On 12 October Meta introduced Muse Wonder and Muse Wonder Full, “our most capable models to date”.
- They are built for agents: long tasks such as handling an insurance claim or building an app.
- Muse Wonder is available now on the Meta Model API and in Muse Code; Wonder Full only for Muse Code subscribers.
- The Muse agent will use it soon, starting with subscribers.
⚠ Pending verification. We have not yet confirmed these details with an official source. Check them before relying on them:
- API pricing, context window size and benchmark results: the announcement gives no figures.
- How Wonder and Wonder Full differ beyond where they are available.
- Whether they will reach the European Union: Meta names no countries.
A month after launching Muse, its personal agent, Meta has introduced the models that will power it from now on: Muse Wonder and Muse Wonder Full. Meta calls them “our most capable models to date” and says they build on what it learned with Muse Spark, the model it used until now.
What they are for
These are not chat models but models for agents: systems that carry out multi-step tasks for you. Meta gives two examples, handling an insurance claim or building an app. For that it highlights two abilities:
- Active Memory: it carries context across conversations, so you don’t have to explain again who you are or where you left off (what the glossary calls agent memory).
- Active Focus: it sticks with hard tasks, asks for clarification when something is unclear and checks in with you before important actions, such as paying for or sending something.
Where you can use them
| Model | Available in |
|---|---|
| Muse Wonder | Meta Model API (for developers) and Muse Code, Meta’s coding agent |
| Muse Wonder Full | Muse Code, subscribers only |
| In the Muse agent | “Soon”, subscribers first |
What the announcement leaves out
Meta gives no prices, no context window size and no benchmark results, and it does not explain how the Full version differs from the standard one. It only claims state-of-the-art performance on agentic tasks, without numbers. It says a full safety report will be available soon. Until then, treat comparisons with caution: the other companies publish their results when they launch a model.
What it means for you
- If you use the Muse agent (for now only in the United States), you will notice the change when Meta switches it on, starting with paid plans.
- If you code, you can try Muse Wonder in Muse Code or on the API and compare it with the models in the comparison, where we have not added it yet because Meta has not published its price.
- If you worry about an agent acting on its own, the Active Focus idea points the right way: ask before acting. We explain why in agent safety.
