Meta Launches Muse Wonder: API Pricing, Coding Claims, and the Catch With Wonder Full

Meta [introduced Muse Wonder and Muse Wonder Full on October 12](https://about.fb.com/news/2026/10/introducing-muse-wonder-models/), positioning the pair for coding, complex work and longer-running agents. The immediate distinction for developers is **access**: `muse-wonder-1.0` is available in the Meta Model API and Muse Code, while **Muse Wonder Full is not available through the API**. Meta offers Full only through qualifying Muse Code subscriptions. [Source: Meta model documentation](https://dev.meta.ai/docs/models)

What can the new model do?

Meta describes Muse Wonder as a model for multi-file software changes, tool use, sub-agent delegation and recovery from errors during longer tasks. It accepts text, image, video, audio and PDF input, returns text, and has a **1,048,576-token context window**. The company highlights “Active Memory” for carrying relevant details across conversations and “Active Focus” for persisting through blockers, seeking clarification and checking before consequential actions. These are Meta’s product descriptions, not a guarantee that an agent will complete every project unattended. [Sources: Meta research](https://research.meta.ai/blog/muse-wonder); [model documentation](https://dev.meta.ai/docs/models)

Price and availability

For Muse Wonder in **Standard mode** on the Meta Model API, Meta lists **$4 per million input tokens**, **$0.20 per million cached-input tokens**, and **$20 per million output tokens**. Its optional web-search grounding costs an additional **$2.50 per 1,000 queries**. The ordinary Wonder variant is also available in Muse Code. [Source: Meta pricing](https://dev.meta.ai/docs/pricing-rate-limits)

Wonder Full is a different access proposition: Meta lists it for Muse Code’s **Power Usage plan at $50/month** and **Maximum Usage plan at $200/month**, rather than per-token API billing. These plans have usage allowances, so “included” does not mean unlimited. Meta says turning Wonder’s reasoning effort up to `max` in the API **does not** select Wonder Full. Wonder is due to reach the Muse consumer agent later, initially for subscribers; that is not part of today’s availability. [Sources: Meta model documentation](https://dev.meta.ai/docs/models); [Muse Wonder product page](https://dev.meta.ai/models/muse-wonder); [launch announcement](https://about.fb.com/news/2026/10/introducing-muse-wonder-models/)

How strong are the coding results?

Meta reports strong performance on coding and agent evaluations, including long-horizon work. But its [model card](https://research.meta.ai/static/muse-wonder-model-card) is clear about attribution: the comparison set combines internal runs, independent third-party scores and provider-reported competitor scores. For SWE-Marathon, competing results use Codex and Claude Code respectively, so Meta says that row is **not a controlled common-harness comparison**. Treat the charts as evidence about the tested setups—not proof that Wonder Full will outperform another model in your repository. sAIgnals has not run an independent benchmark. [Sources: Meta research](https://research.meta.ai/blog/muse-wonder); [Meta model card](https://research.meta.ai/static/muse-wonder-model-card)

Two catches before trying it

First, **Wonder is hosted, not open-weight**. Meta’s open-weight line is Muse Glimmer; the Wonder variants have the availability described above. Second, review the data-use mode. Meta says Standard-mode API content is not used to train its models, and Wonder has no Contributor-mode API rate. In Muse Code subscriptions, Contributor mode is on by default and permits training use of submitted content in exchange for more usage; subscribers can turn it off. That setting matters for private code and documents. [Sources: Meta models](https://dev.meta.ai/docs/models); [Meta pricing](https://dev.meta.ai/docs/pricing-rate-limits)

**Bottom line:** Muse Wonder is a credible new option to evaluate for coding and agent workflows, with clear API rates and a long context window. Wonder Full may be more capable according to Meta, but its subscription-only access makes the products easy to confuse. Compare them using your own tasks, end-to-end cost and data-handling requirements.