How to access Muse Wonder: API pricing and the Full subscription

Choose an API, a coding tool or the planned Muse experience

Meta announced the Wonder family on October 12, 2026 and updated its newsroom page on October 13. Developers integrating a model into their own software can consider Model API; people working on code in a terminal can consider Muse Code. Those are the announced routes, rather than a universal switch across Meta apps.

Wonder Full is currently restricted to Muse Code's Power Usage and Maximum Usage subscriptions. Wonder-powered Muse experiences are described as coming soon, starting with subscribers. No exact launch date or country-by-country arrival schedule is established.

API cost: 1M context is not a free allowance

The official pricing table lists muse-wonder-1.0 with 1M context and USD prices of $4 for input, $0.20 for cached input and $20 for output per million tokens. Context describes how much information a request can accommodate; it is not a free account allowance or proof that every client handles the entire window identically.

For a hypothetical request with 100,000 uncached input tokens and 10,000 output tokens, those two rates give $0.40 plus $0.20, or $0.60. This is a calculation example, not a bill. It excludes tools, other calls, taxes and account terms; cached-input pricing cannot automatically be applied to every input token.

Full's row points to subscription access and leaves API input, cache and output prices blank. Do not apply Wonder's metered prices to Full or assume that appearing in the model table makes Full callable through the API.

What does Maximum's 80x comparison mean?

Josh Walters' developer announcement compares Maximum Usage with the Everyday Usage limit and retains the same five-hour and weekly reset periods. Power Usage also includes Full. The announcement does not provide complete subscription prices or a task-by-task consumption conversion.

Before choosing a plan, inspect your account's available subscriptions, Full selection, current usage and reset time. An 80x limit is not unlimited access, 80x speed or a guarantee that a particular task can run 80 times. No subscription was purchased or private account allowance inspected for this article.

What should an existing client check first?

The developer announcement describes obtaining a Meta API key, pointing an OpenAI-compatible client at Model API and selecting muse-wonder-1.0. Copy the actual base_url, authentication requirements and request example from current official documentation. This article does not invent an endpoint or credential and has made no model call.

A useful first check is a small task with nonsensitive data in an authorized test environment. Verify the returned model, usage, error handling and budget before expanding the work. The official cookbook has individual examples, but some still target Muse Spark; do not assume every earlier recipe validates all Wonder capabilities.

What do Active Memory and Active Focus describe?

Meta describes Active Memory as retaining relevant context across conversations and retrieving it when needed. Active Focus concerns sustained progress, alternative approaches, clarification and confirmation before important actions. These are the publisher's descriptions of training and capability, rather than a measured success rate from PuppyIP.

For a long-task evaluation, retain requirements, permissions, a cost budget and checkable results. One proposed test is to ask for an explanation and repair in an isolated code copy, then inspect the changes and verification results yourself. This is a suggested scenario, not a project installed or completed by PuppyIP.

Expanded global access still requires eligibility checks

The API product page labels the service a public preview with expanded global access. That does not establish availability for every country or account. Check service regions, identity requirements, terms, plans and fees at the official entry point. A complete country list was not established in this review.

The detailed Muse Code page required login during this public review, so account-specific prices and controls could not be confirmed there. Public launch and pricing facts help you choose a route; personal availability still needs a separate check. A different network exit does not create subscription or regional eligibility.