Muse Wonder price: API rates and worked budgets

Which Muse Wonder price applies?

The direct API price table applies to muse-wonder-1.0 in Standard mode. Meta does not offer a Contributor-mode API discount for this model. The table also specifies the same token rates across short and long contexts.

Output is five times the fresh-input price per token. For a workflow that produces long reports or repeated code revisions, the response budget deserves as much attention as prompt length. A small prompt does not guarantee a small bill.

These rates belong to ordinary Muse Wonder. Muse Wonder Full uses a separate subscription access route. A subscription fee and an API token price answer different budgeting questions.

Context capacity is different from a spending allowance

Meta's model documentation gives Muse Wonder a 1,048,576-token context window and lists text, image, video, audio and PDF inputs with text output. A large context window lets a request carry more material; it does not include a prepaid token allowance.

For a document workflow, compare the tokens needed for the relevant passages with those needed for the entire archive. Preserve enough context for a correct answer, then measure the difference. Removing evidence merely to reduce a prompt can make the final result more expensive if it causes failed runs and manual repair.

A request budget you can reproduce

Count fresh input, cache-hit input and billable output separately. For token counts in individual tokens, use:

USD = (fresh input × 4 + cached input × 0.20 + output × 20) / 1,000,000

Suppose every request sends 10,000 fresh input tokens and produces 2,000 output tokens. Input costs $0.04 and output costs another $0.04. One thousand identical requests cost $80. At 10,000 requests the same assumption produces an $800 token budget.

The rows hold usage constant within each example. They say nothing about whether a coding assignment finishes in one request. If an agent makes eight calls, add all eight calls. Keeping only the final answer's token count leaves earlier attempts out of the budget.

How cache hits change the total

Meta tells developers to check the response's cached_tokens count. Only tokens billed as cache hits receive the lower rate. Repeating text in your own prompt is not evidence that all of it received cached billing.

In the second row, $0.008 buys 2,000 fresh tokens, $0.0016 buys 8,000 cached tokens and $0.04 buys the output. The total falls from $0.08 to $0.0496, a 38% saving. The cached-input rate itself is 95% lower than fresh input, but that discount does not apply to the output portion.

If your usage export includes cached input within total input, subtract cached tokens before calculating the fresh-input charge. Charging the full input count at $4 and then adding a cache charge would count those tokens twice. Use the categories returned by your endpoint and retain a sample usage record alongside the calculation.

Add web search as a separate line

Meta lists web search grounding at $2.50 per 1,000 search queries, additional to token charges. Two search queries add $0.005 to a request. Applied to the cached example above, that gives $0.0546 per request, or $54.60 for 1,000 requests.

Count actual search queries rather than assuming one query per user message. Also reserve separate lines for your own hosting, storage and any other services. The totals here cover only the stated Meta usage; they are not a tax-inclusive checkout quote.

Check cost per accepted result

Choose a small set of tasks from your own workload and write down the acceptance conditions before running them. Record the total bill, the number of accepted results and the time spent correcting mistakes. Divide the bill by accepted results. For example, an $8 trial yielding 16 accepted results costs $0.50 per accepted result, even if it made many more API requests.

Meta's billing documentation describes charges accumulating with usage and being collected at a payment threshold or the monthly billing date. A payment threshold is a collection mechanism; set your application's spending controls separately.

AI API Prices checked the linked official documents on October 14, 2026. We have not run a paid performance comparison for this article. For a coding subscription decision, continue to the Muse Wonder Full plan guide; for other providers, browse the model price catalogue.