Meta released Muse Wonder, its new AI model for personal-agent and coding work, on October 12. The practical question is which version your app or plan actually includes.
Regular Wonder is available through Muse Code and Meta’s developer API. Wonder Full needs a Power Usage or Maximum Usage Code plan. For the consumer Muse app, check the offer in your account: Meta’s launch statements differ on timing.
Here, Model API means the developer connection for using Wonder inside your own software. Muse Code is Meta’s command-line coding tool. The consumer Muse app is the personal agent on a phone or web browser. The same model name crosses all three, but the billing and access rules do not.

If you use the Muse app, what should you check?
The Muse app is Meta’s personal agent for everyday tasks. Wonder is the new model that can power it. Zuckerberg’s launch post says subscribing can power your Muse with Wonder today; Meta’s newsroom says Wonder will reach more Muse experiences in a subscriber-first rollout. Those statements leave the exact timing for any one account unclear.
If you want Wonder in the app, open the consumer Muse subscription offer in your own account and look for explicit Wonder access and timing. If the offer or account does not confirm it, wait for that confirmation rather than assuming the announcement changed your account.
Do not use an AI assistant’s answer to “what model are you?” as proof of your entitlement. A plan description or official account notice is a stronger checkpoint. Meta has not given a universal date for everyone in the consumer app. And a Muse Code Power Usage plan, despite its similar name, is for the coding product; it is not documented as an upgrade to the consumer Muse subscription.
If you are deciding what to let the current Muse app do for your business, my guide to its first small-business tasks and approvals covers that separate workflow decision.
Which Muse Code plan gets Wonder Full?
Meta’s current model list says regular Muse Wonder is available in Muse Code with pay-as-you-go billing and on every Code subscription tier. Wonder Full is limited to Power Usage and Maximum Usage subscriptions. It is a separate model: turning regular Wonder’s reasoning setting up to “max” does not turn it into Full. Plan benefits may vary by region, so check the exact offer for your account before changing plans.

Before sending private work: Meta says Muse Code subscriptions start in Contributor mode, which permits submitted prompts, code, files, context and responses to be used to train future Meta models in exchange for more usage. In Standard mode, Meta says that content is not used for model training. You can turn Contributor off in Muse Code settings for future content, with a lower usage allowance. Check the setting and current terms before sending confidential material.
A Code subscription covers its command-line tool credential. Meta says an additional API key under the same developer account is billed separately on pay-as-you-go terms. If you need Full for a coding workload, compare the plan’s actual price, allowance and mode in your account with the work you need it to complete. A benchmark lead alone will not tell you that.
What does regular Wonder cost through the API?
For developers, the callable model is muse-wonder-1.0. Meta’s October 13 rate card lists $4 per million uncached input tokens and $20 per million output tokens in Standard mode; cached input is $0.20 per million when the request actually gets a cache hit. Wonder Full has no general Model API route. Meta also says regular Wonder on that API is Standard mode only.
For a made-up billing example, not an observed run, 100,000 uncached input tokens plus 10,000 output tokens would cost $0.60 with Wonder at those rates. The same token counts with Meta’s cheaper Spark 1.3 Standard rate would cost about $0.17. That excludes tool calls, web searches, retries and caching, and equal token counts do not mean equal task quality. I would test one real workload and track both the result and bill before switching a production agent.
What is actually new beyond the access rules?
Meta describes Wonder’s Active Memory as retrieving relevant details from earlier conversations, and Active Focus as continuing through obstacles, trying other approaches and checking before important actions. Those are useful ambitions for a personal agent or a long coding task. They are Meta’s descriptions, not a test I ran or a promise that every Muse account gets the same behavior today.
My next step depends on the product: in Muse, verify the subscription offer and rollout on your account; in Muse Code, choose the model and inspect both plan and Contributor setting; in an API integration, test regular Wonder’s output and metered cost against the model you already use. If your current setup handles your work, keeping it is a perfectly sound decision.
Sources and corrections
I checked Meta’s October 12 model announcement, newsroom summary, Zuckerberg’s consumer subscription statement, model-access table, Code subscription terms and API pricing on October 13, 2026. The consumer statements leave account-level timing unresolved; this article does not claim firsthand access or comparative performance. Send corrections to hello@theprotoolkit.com.