Muse Wonder
Muse Wonder is Meta's October 12, 2026 frontier agent model, available as muse-wonder-1.0 in Meta Model API and Muse Code. The API offers Standard mode with a 1,048,576-token context; Muse Wonder Full is a different, subscription-only variant.
The decision is which execution surface, privacy mode and task budget fit the job. A high reasoning effort on Wonder does not select Full, and developer availability does not imply Wonder has reached the consumer Muse app.
Official launch and current provider documentation support identity, access and specification claims. Publisher performance claims are attributed; no independent model call or task-quality test was run.
- Multi-file software engineering
- Long-running tool workflows
- Personal/professional coordination with reviewed permissions
Meta documents text, image, video, audio and PDF input with text output and a 1,048,576-token context window. Standard API rates are $4 input, $0.20 cached input and $20 output per million tokens. Use muse-wonder-1.0 on the documented Responses, Chat Completions or Messages interface; check the selected interface for request details.
Muse Wonder is available through Model API and every Muse Code plan. Wonder Full is unavailable in Model API and requires Muse Code Power Usage or Maximum Usage. Setting reasoning effort to max on Wonder changes effort, not model identity.
Standard mode does not use submitted content to train Meta models. Muse Code subscribers can use Contributor mode, on by default, which provides more usage in exchange for allowing training on submitted content; it can be disabled. Wonder pay-as-you-go does not offer Contributor mode. Review mode and permissions before sending private work.
Meta describes carrying useful context across conversations and recovering from errors during long tasks. Its showcase includes planning, game development and coordination examples. These are publisher demonstrations with external tools, not independent proof of unattended success, privacy or a single unassisted model call.
Meta positions Wonder for complex, long-running work and Spark for cheaper high-volume loops. Start with a repository or planning task with explicit acceptance criteria. Inspect changes, failed steps, tool permissions, token costs and completion before expanding the task. Vendor benchmark rows use different harnesses and effort settings, so they do not establish universal superiority.
Muse Wonder FAQ
Common questions about Muse Wonder.
Is Muse Wonder Full available in the API?+
No. Meta lists Full only in qualifying Muse Code subscriptions. The public API model muse-wonder-1.0 is Wonder.
Does max effort enable Wonder Full?+
No. Reasoning effort changes how Wonder works; Full is a distinct model selected through an eligible Muse Code subscription.
Are all Muse users already on Wonder?+
No such universal rollout is established. Meta says Wonder will reach Muse later, first for subscribers.
Does Meta train on my code?+
It depends on the mode. Standard mode excludes submitted content from training; Muse Code Contributor mode allows training in exchange for more usage and is enabled by default for subscribers.