Meta launched Muse Wonder and Muse Wonder Full on October 12, 2026. The new AI models focus on tasks that require planning, remembering relevant context and working through multiple steps. Muse Wonder is available through Meta Model API and Muse Code; Muse Wonder Full has a separate subscription-based access route. Meta also announced a future rollout into its Muse personal agent. See the official launch announcement.
This guide explains the features, access options and pricing, with details checked on October 14, 2026. It distinguishes Meta's published capabilities from suggested ways to evaluate them.
What is Meta Muse Wonder?
Muse Wonder is an AI model series designed for personal and professional agents. An agent uses a model alongside tools to pursue a goal: planning steps, acting on information and checking progress. The model's capability and the permissions of the surrounding application both affect what it can do.
Meta's research overview identifies Active Memory and Active Focus as central capabilities, alongside coding. These names describe remembering useful context and maintaining effort on a task, respectively.
Muse Wonder at a glance
| Detail | What Meta has announced |
|---|---|
| Launch date | October 12, 2026 |
| Variants | Muse Wonder and Muse Wonder Full |
| Standard model ID | muse-wonder-1.0 |
| Context | Marketed as 1M tokens |
| Developer access | Meta Model API and Muse Code |
| Full variant | Muse Code Power Usage and Maximum Usage subscriptions |
| Consumer rollout | Muse experiences coming later, subscribers first |
Sources: Meta Newsroom, model specifications and model availability documentation.
Key Muse Wonder features explained
Active Memory: recalling relevant context
Meta describes Active Memory as carrying important information across conversations and retrieving it when useful. For example, a project assistant needs to recognize when an earlier requirement affects a new request. That example illustrates the intended benefit; it is not a result from our own testing.
Persistent memory also raises a product-design question: how will a user inspect, correct or remove an outdated detail? Treat that as part of an evaluation of the complete assistant experience. Source: Meta's research explanation.
Active Focus: working through difficult tasks
Active Focus describes sustained work toward a goal, including trying other approaches, recovering from errors and asking for clarification. Meta says the models should check in before important actions. Its research overview describes FocusBench and SWE-Marathon as evaluations of focus and sustained software work.
A useful test is to introduce a realistic blocker, such as missing information, and inspect the response. Does the agent ask a relevant question and preserve the task's constraints? Source: Muse Wonder research overview.
Coding, tool use and computer interaction
The developer launch post emphasizes extended reasoning, tool invocation, failure recovery and end-to-end coding tasks. Meta's model page also describes computer use and changes across large repositories.
For a development team, a practical trial would be a small feature that touches several files. Define acceptance criteria, supply the existing tests and review the resulting diff. Record whether the system completes the feature, creates regressions or needs repeated corrections. A generated app is only useful if its behavior meets the requirements.
Muse Wonder vs Muse Wonder Full
The main access difference is straightforward:
| Variant | Meta Model API | Muse Code |
|---|---|---|
| Muse Wonder | Available in Standard mode | Pay-as-you-go and subscription plans |
| Muse Wonder Full | Not available through Model API | Power Usage and Maximum Usage plans |
Meta positions Full as its higher-capability variant. Increasing reasoning effort on the standard API model does not switch it to Full. Check the model availability documentation before choosing an integration.
Muse Wonder pricing and context window
Meta lists these Standard-mode API rates for muse-wonder-1.0, in US dollars per million tokens:
| Token type | Price |
|---|---|
| Input | $4.00 |
| Cached input | $0.20 |
| Output | $20.00 |
The pricing documentation also lists web-search grounding as an additional charge. These token rates are separate from Muse Code subscription fees and do not describe a pay-as-you-go API price for Wonder Full.
For a simple estimate, 100,000 uncached input tokens and 10,000 output tokens cost $0.60 at those rates: $0.40 for input plus $0.20 for output. This hypothetical calculation excludes extra tools, retries and taxes.
The API overview specifies a 1,048,576-token context window and currently labels Muse Wonder as a preview. Context size describes capacity; it does not establish the accuracy of a response to a long document. Check current documentation before building against a preview release.
How to access Muse Wonder
- Open Meta's developer model page and choose API access or Muse Code.
- For an API integration, follow the official quickstart documentation and select
muse-wonder-1.0. - For Wonder Full, check the eligible Muse Code subscription plans.
- Review data-use settings before sending private project material.
That last step matters: Meta's subscription documentation says Muse Code subscriptions default to Contributor mode, which allows content to be used to train future models in exchange for more usage. Users can turn it off. The model documentation distinguishes this from Standard mode, where content is not used for model training.
Muse Wonder vs Gemini 4 Argon: what should you compare?
Both announcements emphasize complex workflows, but choosing a model requires more than comparing launch descriptions. Use the same task, input material and acceptance criteria, then measure correctness, time, cost and the amount of human correction needed.
For Google's inputs, context limits and access routes, read our Gemini 4 Argon features and capabilities guide. We have not run a head-to-head benchmark of these models.
Muse Wonder frequently asked questions
Is Muse Wonder free?
Meta publishes paid API rates for Muse Wonder. Muse Wonder Full requires an eligible Muse Code subscription. Check the current developer pricing before assuming a free access route.
Is Muse Wonder Full available through the API?
Meta's model documentation lists Full as available through Muse Code's Power Usage and Maximum Usage subscriptions, rather than Meta Model API.
Is Muse Wonder already available in the Muse app?
The launch announcement says Muse Wonder will power Muse experiences soon, with subscribers receiving it first. Developer availability should not be treated as confirmation that every consumer account already has access.
What is the difference between Muse Wonder and Muse Spark?
Meta's documentation positions Wonder for longer, more complex tasks and Spark for workloads where latency or cost matters more. Test the workload you intend to run before selecting either model.
Sources for these access answers: Meta's launch announcement, model documentation and pricing. Cover image: an original AI-generated concept illustration, not an official Meta asset.