Muse Wonder Pricing: Pay in Dollars or Pay in Data

The short version: Muse Wonder pricing comes in two currencies. Pay in dollars and Meta charges $4 per million tokens in and $20 out, exactly what Claude Opus 5.5 costs, and promises not to train on your work. Pay in data and Meta’s own price list shows what that is worth to it: on its smaller Muse Spark model, letting Meta train on your content cuts the bill by 92 to 95 percent. Meta’s terms also close that cheaper lane to most real business work, because they forbid sending personal or confidential information through it.

That is the useful thing to know about Meta’s new frontier model, released October 12, and it is written plainly in Meta’s developer documentation rather than hidden. Below, the two price lists side by side, what the choice is worth to a small office in actual dollars, and the one setting to check if anyone on your team uses Meta’s coding tool.

What is Muse Wonder, and how good is it?

Muse Wonder and Muse Wonder Full are the first models from Meta Superintelligence Labs that Meta itself calls state of the art for “personal and professional agentic tasks.” Meta pitches two new abilities: Active Memory, which carries context about you across conversations, and Active Focus, which keeps working through blockers and checks in before important actions. Meta’s examples include explaining a lease and working through financial or health information.

On the independent Artificial Analysis Intelligence Index, Muse Wonder scores 52, ninth of 216 models tracked, as read on October 13. VentureBeat’s launch report puts Claude Opus 5.5 at 58 and the larger Wonder Full at 56 on the same index. So Meta is now genuinely in the top group, which is good news for anyone who buys AI: at least four American labs competing at the frontier is how prices stay where they are.

One practical note before anyone wires it into a customer chat. At its maximum reasoning setting, Artificial Analysis measured 36 seconds before Muse Wonder starts answering, against a median of about six seconds. That is fine for a report you will read after lunch. It is not fine for a customer waiting on your website. Meta also labels Wonder “in preview.”

Muse Wonder pricing: the two price lists

Meta sells its models in two modes, set out on its pricing and rate limits page. In Standard mode, its Model API terms say in section 5.1 that Meta “will not use Content submitted in a request to the Standard Services to train Meta Models.” In Contributor mode, section 6.1 says the opposite: Meta may use your content to train and improve its models. Here are the published rates per million tokens, read on October 13:

Read the Spark rows as an exchange rate. Letting Meta train on your content takes 92 percent off input, 95 percent off output and 99 percent off cached input. Few vendors print the value of a customer’s data this openly, and it is worth a moment, because it tells you how much Meta wants real-world work to learn from. On the pay-as-you-go API, Wonder itself is sold in Standard mode only, so if you call it directly, your content is not training material.

What is the discount worth to a small business?

Less than it sounds. Take a busy month for a small office: one million tokens read and 200,000 written. That is very roughly 750,000 words of documents and email in and 150,000 words of drafts out, more than most five-person shops would push through an API.

• Muse Spark, Standard: $1.25 for input plus $0.85 for output, $2.10 a month.
• Muse Spark, Contributor: $0.10 plus $0.04, $0.14 a month.
• Muse Wonder, Standard, for comparison: $4.00 plus $4.00, $8.00 a month.

The data discount on that heavy month is $1.96, or about $24 a year. For a small business the money is close to irrelevant either way. The decision comes down to what is in the content, and Meta’s terms answer that for you.

Why most business work cannot use Contributor mode anyway

Section 6.2 of Meta’s Model API terms says: “You must not submit sensitive, confidential, or personal information to the Contributor Services.” The same section says Contributor mode has no way to exclude specific traffic from training; if something must not be trained on, it goes through Standard.

Think about what a small business actually asks an AI to do. Reply to a customer by name. Summarize a supplier contract. Draft a reminder to a tenant who is behind on rent. Sort job notes that carry home addresses. Nearly all of it contains personal or confidential information, which puts it outside the cheap lane by Meta’s own rule. What is left for Contributor mode is genuinely public material: rewriting your own published web copy, brainstorming blog topics, drafting a generic flyer. That is a real use, and at those prices a nearly free one, but it is a narrow one.

Where Contributor mode is on by default

There is one place the choice is made for you. Wonder Full, the stronger of the two models, is not on the API at all. Meta’s model documentation says it is “available only in Muse Code with a subscription,” on the Power Usage and Maximum Usage plans. VentureBeat reports those plans at $50 and $200 a month; Meta’s subscription page lists the plans and their usage multiples but, when we read it on October 13, not their prices.

Meta’s subscription page is direct about the default: “Muse Code subscriptions default to Contributor mode.” Meta’s documentation says the mode gives subscribers more usage in exchange, and that it can be turned off in Muse Code’s settings at any time. Section 3.3 of the terms adds the detail that matters: switching to Standard “would apply prospectively only,” and Meta’s rights over content already sent survive the change.

So if you, a partner or a freelancer builds anything for your business in Muse Code, whether that is a booking page, a quote calculator or a script that tidies your customer list, turn Contributor mode off before the first file goes in, not after. The order matters more than the setting.

Three habits that follow

Match the lane to the content, not the price. Public material can ride the discount; anything with a name, address, account or contract in it goes through Standard mode, at any vendor. The Spark example shows the gap is a couple of dollars a month.

Check the default before the first upload. Settings that only change the future need to be set at the start. That applies to Muse Code and to any AI tool your team signs up for on a free or discounted plan.

Treat Wonder as a back-office worker for now. A 36-second start is a poor fit for a live customer chat and a fine fit for overnight document work. If you use Meta’s Muse agent, its approval controls are worth setting carefully, and if your concern is keeping sensitive work entirely in-house, Meta’s free Muse Glimmer model runs on hardware you own. For a price comparison across the other frontier labs, see Opus 5.5 against GPT-6 Sol.

Frequently Asked Questions

How much does Muse Wonder cost?

On Meta’s Model API, Muse Wonder costs $4 per million input tokens, $20 per million output tokens and $0.20 per million cached input tokens, as listed on Meta’s pricing page on October 13, 2026. That is the same rate card as Claude Opus 5.5. Built-in web search adds $2.50 per 1,000 searches.

Does Meta train on what I send to Muse Wonder?

Not if you use the pay-as-you-go API, which offers Muse Wonder in Standard mode only, and Meta’s terms say Standard content is not used to train its models. Muse Code subscriptions are different: they default to Contributor mode, which allows training, until you turn it off in settings, and the change only applies going forward.

Can I get Muse Wonder Full through the API?

No. Meta’s documentation says Muse Wonder Full is available only inside Muse Code on the Power Usage and Maximum Usage subscription plans. Setting Muse Wonder’s reasoning effort to maximum does not switch it to the Full model.

Is Contributor mode worth it for a small business?

Only for public material. It cuts Muse Spark prices by 92 to 95 percent, but on a heavy month for a small office that is about $2, and Meta’s terms forbid sending personal, sensitive or confidential information through Contributor mode, which rules out most customer, employee and contract work.

Would you trade your business’s data for a discount if the content were genuinely public, or is that a line you would rather not cross at any price?