Muse Wonder

Muse Wonder is Meta’s multimodal model family for complex coding, reasoning, and long-running agent tasks. It includes Muse Wonder, available through Meta Model API and Muse Code, and Muse Wonder Full, a higher-capability variant available only through eligible Muse Code subscriptions. Meta positions both for extended workflows involving planning, tool use, and error recovery.

What It Is

These are models used inside applications and agent tools. Developers provide the tools, execution environment, and agent harness. Meta describes the family as supporting multi-file coding, sub-agent delegation, computer use, and sustained reasoning across complex tasks.

Wonder and Wonder Full

Muse Wonder uses the API model ID muse-wonder-1.0 and supports a 1,048,576-token context window. It accepts text, image, video, audio, and PDF input and returns text. Meta’s API documentation labels it as preview.

Muse Wonder Full is the higher-capability variant, with a context window advertised as 1M tokens. It is available through Muse Code rather than Meta Model API. Setting reasoning effort to max on Muse Wonder does not select Wonder Full.

Developer Access

Meta Model API supports Responses, OpenAI-compatible Chat Completions, and Anthropic-compatible Messages request formats. Developers can configure compatible clients with the Meta endpoint and an API key. Documented capabilities include parallel tool calls, structured output, search grounding, and multimodal understanding.

Muse Code provides a ready-made development environment for using the models in coding workflows. Access to each variant depends on the connected subscription or billing mode; Wonder Full is restricted to eligible subscriptions.

Data Modes

Wonder API requests use Standard mode, where submitted content is not used to train Meta models. Contributor mode is not available for Wonder API requests.

Muse Code subscriptions support Standard and Contributor modes for both variants where available. Contributor mode is enabled by default and permits Meta to use submitted content to train future models; users can disable it in settings. Additional API keys are billed separately from a Muse Code subscription.