Meta Muse Wonder and Wonder Full: the access and data-sharing choices

Muse Wonder’s Model API and Muse Code routes and standard USD list rates, checked 11 November 2026. Web search is charged separately. Original editorial graphic by PING! Magazine. Sources: Meta developer documentation and API terms.

A new model name is only the beginning of a buying decision. Meta introduced Muse Wonder and Muse Wonder Full on 12 October, promising agents that remember context and persist through difficult work. For someone considering them for a software project, the immediate questions are more concrete: which version can you actually use, how is it billed, and what happens to the code and documents you send?

Three things to know

• Wonder is offered through Meta Model API and Muse Code; Wonder Full is a separate variant restricted to specified Muse Code subscription plans.
• A higher reasoning-effort setting does not switch standard Wonder into Wonder Full.
• Meta says Muse Code subscriptions default to Contributor mode, which allows submitted content to be used for model training. Check that choice before sending work.

Developer access and the consumer rollout are separate

Meta’s 12 October announcement said Wonder was available through Model API and Muse Code, with wider Muse experiences planned for subscribers first. On 6 November, Meta said it was rolling out specialised versions of existing models trained using Muse Wonder in Muse and Meta AI. Access to those consumer features is a separate question from access to the developer variants. PING has not checked a UK, Kenyan or Nigerian account.

Wonder Full is a different choice, not just more thinking time

Meta’s current model guide lists Wonder in Standard mode on Model API and on all Muse Code plans. It lists Wonder Full only in Muse Code’s Power Usage and Maximum Usage plans. Raising Wonder’s reasoning effort to max leaves the model as Wonder. Record the variant and route before comparing results.

More usage comes with a data-use decision

The model catalogue says that Standard-mode content is not used to train Meta models. It describes Contributor mode as exchanging permission to use submitted content for more usage, with Muse Code subscriptions defaulting to that mode. The description includes prompts, code, files, context and responses, not merely feedback scores. Meta says it can be turned off in settings. Those are provider descriptions, not PING's independent privacy audit.

This matters for an agency handling a client's website or a developer working on an employer's repository. Decide what information you are authorised to share before connecting a work folder. A personal subscription does not establish permission from a client or employer. For a first evaluation, PING suggests fictional requirements and a disposable project without credentials, customer records or confidential documents. This is a proposed precaution, not a trial we have conducted.

An API rate is not the price of finishing your task

Meta’s pricing page, rechecked on 11 November, lists Muse Wonder at US$4 per million input tokens, US$0.20 for cached input and US$20 for output. Web search grounding costs an additional US$2.50 per 1,000 queries. These are Model API list rates. A long agent run can contain several requests and tool actions, so track the total usage for the task.

Memory and persistence need a task-level test

Meta's research account describes Active Memory as carrying relevant context across conversations, and Active Focus as sustaining work, trying alternatives and checking in before important actions. These are the company's capability claims. They do not prove that your chosen application remembers the right information, stops at your permission boundary or finishes a particular job correctly. PING has not reproduced the company's evaluations.

Give a first trial a clear finish line

Choose a small, disposable task with an observable result: for example, turn fictional event details into a page whose dates, links and layout can be checked. Write the expected output and forbidden actions before beginning. Keep the model variant, data-use mode and billing route in the record. Review changes and tool actions, then note what needed human correction. Compare the completed result and total recorded usage, not just the first impressive response. Stop if the account, charging route or sharing permission is unclear.

The useful question for a team in Lagos, Nairobi or London is whether the exact setup fits its work and information boundaries. Wonder's launch is a reason to evaluate that setup, not to hand over an entire project on the strength of a chart. Follow more documented software and agent coverage on PING's AI desk.