Meta logo with its blue infinity-shaped mark and the Meta wordmark, the company behind the Muse Wonder AI models

Muse Wonder: Power-User Guide to Meta’s Model API, Pricing and Muse Code

by

Last updated: October 30, 2026

Muse Wonder is Meta’s newest AI model, released on October 12, 2026. Developers can call it as muse-wonder-1.0 on the Meta Model API for $4 per million input tokens and $20 per million output tokens. The larger Muse Wonder Full runs only in Muse Code on the top two subscription plans.

Meta announced both models in its Muse Wonder newsroom post, describing them as its most capable models so far. This guide covers what is new, how to call the API, what it costs, how reasoning effort works, and how to use the models in the Muse Code command-line tool.

Contents

What’s new in Muse Wonder

Meta released two models on the same day. Per the Meta Model API models page, Muse Wonder takes text, image, video, audio and PDF input, returns text, and has a 1,048,576-token context window. Muse Wonder Full is the larger model, reserved for Muse Code subscribers.

Meta highlights these capabilities:

  • Active Memory: carries what matters across conversations and brings back details relevant to the current task.
  • Active Focus: keeps working when a task gets hard, tries other approaches, recovers from errors, and checks in before important actions.
  • Agent tools: sub-agents, parallel tool calls, computer use, structured output and web search grounding.
  • Coding: Meta reports that Muse Wonder Full tops the DeepSWE and SWE-Marathon coding benchmarks in independent third-party runs. These are vendor-reported claims.

Meta says the models were trained at its Prometheus data center and that a full report with additional safety evaluations will be available soon.

Meta’s earlier models are covered in our posts What is Meta Muse Spark 1.3? and What is Muse Spark Open?.

Muse Wonder models, pricing and limits

The figures below come from Meta’s pricing and rate limits page and models page.

Feature Muse Wonder Muse Wonder Full
Where available Meta Model API and Muse Code Muse Code only
Model ID muse-wonder-1.0 Selected in Muse Code
Input price $4.00 per 1M tokens Included in subscription
Cached input $0.20 per 1M tokens Included in subscription
Output price $20.00 per 1M tokens Included in subscription
Long-context premium None Not applicable
Subscription plans Every Muse Code plan Power Usage and Maximum Usage
Context window 1,048,576 tokens Not listed in this guide’s sources

On the API, Muse Wonder allows 3,000 requests per minute and 4,000,000 tokens per minute. These limits apply to your whole team, not to each API key. Web search grounding is billed at $2.50 per 1,000 queries. Contributor mode, which trades training rights for a lower price on some Meta models, is not available for Muse Wonder on pay-as-you-go.

How to make your first Muse Wonder API request

The Meta Model API is compatible with common SDKs, so most existing code needs only a new base URL, key and model ID:

  1. Create an API key in the Meta developer platform and export it as MODEL_API_KEY.
  2. Point your client at https://api.meta.ai/v1 and send the key as a Bearer token.
  3. Choose an API shape: Responses, Chat Completions or Messages. The OpenAI SDK, the Anthropic SDK and OpenAI-compatible agent tools all work.
  4. Set the model to muse-wonder-1.0 and, if needed, a reasoning effort level.
  5. Watch usage, because reasoning tokens are billed as output at $20 per million.
curl https://api.meta.ai/v1/responses \
  -H "Authorization: Bearer $MODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"muse-wonder-1.0","reasoning":{"effort":"high"},"input":"Summarize the risks in this contract."}'

For other agent-focused releases this month, see our Cursor Agent SDK guide.

Reasoning effort settings

Per Meta’s reasoning documentation, muse-wonder-1.0 accepts six effort levels: minimal, low, medium, high, xhigh and max. Sending none returns an HTTP 400 error.

  • Reasoning tokens count toward max_output_tokens and are billed as output, so raise that limit when you raise effort.
  • The reasoning.summary field accepts auto, concise or detailed.
  • Setting effort to max does not route the request to Muse Wonder Full.

Using Muse Wonder in Muse Code

Muse Code is Meta’s command-line coding agent. Install it with curl -fsSL https://dev.meta.ai/install.sh | sh on macOS or Linux, or irm https://dev.meta.ai/install.ps1 | iex in Windows PowerShell, then run muse. Per the Muse Code configuration page, you can pick a model with --model or the /models command, and change effort with --reasoning-effort or /effort.

The Muse Code subscriptions page lists four plans. Everyday Usage allows 10 to 50 prompts every five hours and includes Muse Spark, Muse Wonder and Muse Image. High Usage gives 5 times that usage, Power Usage 20 times and Maximum Usage 80 times, and only the last two include Muse Wonder Full. You subscribe through Accounts Center, which /upgrade opens from a session.

Contributor mode is on by default for subscribers, giving more usage in exchange for letting Meta train on your content. Turn it off in /settings, or for one run with --contributor-mode off. A subscription covers Muse Code only; extra API keys are billed pay-as-you-go.

For competing models released this month, see our Gemini 4 Argon guide and Kimi K3.1 Pro guide.

How Muse Wonder compares with Muse Spark

Muse Spark 1.3 remains Meta’s lower-cost model at $1.25 per million input tokens, $0.15 cached and $4.25 output. Muse Wonder costs about 3 times more for input and about 5 times more for output. Meta recommends Muse Wonder for long, complex tasks and Muse Spark for high-volume, cost-sensitive work, and switching is a model ID change.

For another large model released this month, see our Mistral Large 4 guide.

Limitations

  • Muse Wonder is in preview, so behavior and pricing may change.
  • Neither model is currently available in South Korea.
  • Muse Wonder Full has no API access.
  • Output is text only, even though input can include images, video, audio and PDFs.
  • The full safety report had not been published at launch.

Frequently asked questions

What is Muse Wonder?

Muse Wonder is a Meta AI model released on October 12, 2026, alongside a larger Muse Wonder Full. It accepts text, image, video, audio and PDF input, returns text, and has a 1,048,576-token context window. Meta recommends it for long, complex tasks such as coding.

How much does Muse Wonder cost on the Meta Model API?

Meta lists muse-wonder-1.0 at $4.00 per million input tokens, $0.20 per million cached input tokens and $20.00 per million output tokens, with no extra charge for long context. Web search grounding costs $2.50 per 1,000 queries.

Can I use Muse Wonder Full through the API?

No. Muse Wonder Full is available only in Muse Code, on the Power Usage and Maximum Usage subscription plans. Setting reasoning effort to max on muse-wonder-1.0 does not switch the request to Muse Wonder Full.

Does Muse Wonder work with the OpenAI SDK?

Yes. The Meta Model API at https://api.meta.ai/v1 supports the Responses, Chat Completions and Messages APIs, so the OpenAI SDK, the Anthropic SDK and OpenAI-compatible agent tools work after you change the base URL, API key and model ID.

Is Muse Wonder available everywhere?

Not yet. Meta says Muse Wonder and Muse Wonder Full are not currently available in South Korea, on the Model API or in Muse Code. Muse Wonder is also labeled as a preview model in Meta’s documentation.

Should I use Muse Wonder or Muse Spark?

Meta recommends Muse Wonder for long, complex tasks and Muse Spark for high-volume, cost-sensitive work. Muse Spark 1.3 costs $1.25 per million input tokens and $4.25 per million output tokens, and switching between the two is a model ID change.

Sources

Image: Meta Platforms logo, public domain, via Wikimedia Commons. Placed on a white background and resized. Meta is a trademark of Meta Platforms, Inc.



Related answers

Leave a Comment