Meta’s preview reasoning model for demanding coding, analysis and multimodal understanding, with a 1M-token context and six API reasoning-effort levels.
Tool support requires the appropriate API integration; a supported tool is not automatically active in every chat.
Supported API features and tools
Meta documents text, image, video, audio and PDF understanding with text output on the provider API
A 1,048,576-token context window for supplied material and conversation context
Six reasoning-effort settings from minimal through max, with model-determined reasoning when omitted
Provider APIs support streaming, search grounding, tool calling and structured output through compatible integrations
A model designed for complex coding and multi-step agentic work, with execution supplied by the surrounding application
Before you choose
The reviewed Meta references do not establish a standalone maximum output-token allowance or a single knowledge-cutoff date. The 1M context size is not a promise of 1M output tokens or complete recall.
Multimodal understanding produces a text answer. Reading an image, recording or PDF is different from generating an image, audio track or edited document.
Meta lists Muse Wonder on Model API in Standard mode only. Muse Code subscriptions, Contributor mode and Wonder Full have different availability; their terms and capabilities must not be transferred to this API model.
Complex reasoning can still produce incorrect code or miss evidence. Preserve source locations, validate structured results and run acceptance tests before using a proposed change.
Choose the reasoning effort
minimallowmediumhighxhighmax
Meta supports minimal, low, medium, high, xhigh and max. When effort is omitted, the model chooses its reasoning depth; omission does not mean reasoning is off. Chat Completions uses reasoning_effort, while Responses uses reasoning.effort. Compare lower effort for bounded tasks with higher effort for dependent reasoning. Higher effort can increase latency and billed reasoning tokens. Max stays on Muse Wonder, not Wonder Full.
Unsupported settings: none.
The exact model ID is muse-wonder-1.0. Meta provides OpenAI-compatible Chat Completions and a Responses interface; use the documented request format for the chosen endpoint.
Meta Responses exposes the web_search tool and search citations. Read the returned sources and check that each citation supports the associated claim.
Meta counts reasoning tokens toward output-token budgets and bills them at the output rate. A visible answer can be shorter than the total generated token count.
Tool calling and structured outputs are separate API capabilities. A tool workflow must execute authorized calls and return their results; schema-constrained responses still need validation and handling for errors or refusals.
Put it to work
Start with a more useful prompt.
Original examples from EZ Ai Assist. Adapt these to your task and the features available in your workspace.
Workflow 01
Investigate a concurrency defect
Use pasted code and logs to narrow a difficult failure.
Review the worker code and incident logs pasted below. Trace how a queue message could be processed twice after a restart. Rank explanations by the evidence, cite the relevant functions and log entries, and propose a minimal reproduction test for each. Recommend the smallest supported repair with rollback considerations. Separate confirmed observations from hypotheses, and do not claim you ran the code.
Workflow 02
Reconcile a demanding project brief
Turn conflicting requirements into a reviewable decision list.
Compare the three labeled requirement excerpts below. Identify direct contradictions, different assumptions and missing acceptance criteria. Quote the supporting passage for each finding. Propose a decision log with an owner only when the text names one, and list the questions that need a human decision. Finish with a staged implementation outline that does not silently choose between conflicting requirements.
Workflow 03
Review a screen and its error report
Combine an image with a precise written problem statement.
Inspect this screenshot alongside the bug report I provide. Describe only what is visibly supported by the image, then compare it with the expected behavior. Suggest likely implementation areas to inspect without inventing hidden code. Provide a short reproduction checklist and acceptance tests covering keyboard use and a narrow phone screen. State which findings require an interactive test rather than a screenshot.
Developer reference
Meta API pricing
These are published provider API reference prices, not EZ Ai Assist subscription prices or verified customer charges.
Meta Model API · Standard mode · checked October 16, 2026 · USD per 1,000,000 tokens
Token type
Price
Input
$4.00
Cached input
$0.20
Output, including reasoning
$20.00
Additional API fees · USD
Usage
Rate and unit
Web search grounding
$2.50per 1,000 search queries, plus token charges
Meta publishes $4 input, $0.20 cached input and $20 output per million tokens for muse-wonder-1.0. The reviewed table lists no long-context premium and no Contributor API rate for Wonder.
Cache eligibility and the number of search queries affect provider cost. A cached-input rate is not a discount on all input or on generated output.
Meta developer API rates are separate from EZ Ai Assist plans and usage charges. Use the app’s plan and usage information for customer-facing costs.
Common questions
A few things worth knowing.
When should I try Muse Wonder?
Try it for a demanding code review, conflicting requirements or evidence-heavy analysis. Give it a bounded deliverable, source locations and acceptance criteria. Compare the result with a less expensive model on the same task rather than assuming the most capable positioning guarantees a better answer.
Does max reasoning give me Muse Wonder Full?
No. Meta distinguishes muse-wonder-1.0 on Model API from Muse Wonder Full in eligible Muse Code subscriptions. Max changes the reasoning effort on Wonder; it does not change the model into Full or grant a different subscription.
How do I choose reasoning effort?
Meta supports minimal, low, medium, high, xhigh and max. Start with a bounded task and compare the result before increasing effort. When the parameter is omitted, Meta chooses the depth; that is not the same as disabling reasoning. The provider does not support none.
What inputs does Muse Wonder understand?
Meta documents text, images, video, audio and PDFs with text output. Choose clear source material and ask for page, timestamp or passage references where appropriate. Understanding supplied media is not the same as generating new media.
Is the 1M context also an output limit?
No. Meta lists a 1,048,576-token context window, but the reviewed official references do not establish a separate maximum output allowance. Input, reasoning and response budgets are different quantities; a large context does not promise a response of the same size.
How do search and tool calling differ?
Meta Responses provides web_search for grounded retrieval with citations. General tool calling lets a model request an action that the surrounding integration executes and returns. Both need appropriate configuration; a proposed tool call is not itself proof that an action completed.
How should I check a long analysis?
Ask for a bounded deliverable with source locations, explicit assumptions and acceptance criteria. Verify quotations and calculations, test proposed code and review omissions as well as plausible-looking answers. More reasoning does not guarantee correctness.
Are these prices my EZ Ai Assist charges?
No. The table gives dated Meta developer API references. It does not confirm your subscription allowance, final task charge or usage-settlement calculation. Those depend on the application and your plan, which are separate from the provider’s published token rates.
Check the source
Official documentation
Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.