Muse Wonder Pricing: API, Muse Code, and Real Test Costs
Muse Wonder costs $4/$20 per million API tokens. See Muse Code plans, real test costs, and how Polish tokenization affected the results.
👁 115 reads
Muse Wonder costs $4 per million input tokens and $20 per million output tokens in the API - the same rates as Claude Opus 5.5. The more powerful Muse Wonder Full is available only through Muse Code subscriptions costing $50 or $200 per month. But the price list tells only part of the story: in my tests, Polish text used almost half as many tokens with Muse Wonder as with Claude, while the model spent a lot of time “thinking” even on simple questions. Here are the prices and what I actually paid. For an overview of the model, see my Muse Wonder guide.
Prices as of October 12, 2026
- API (Meta Model API and OpenRouter): $4 for input, $20 for output, and $0.20 for cached input per million tokens.
- Muse Code with Wonder Full: Power Usage, $50; Maximum Usage, $200 per month.
- Muse app (US and Canada): $20 and $100 per month; Wonder “coming soon.”
Muse Wonder API pricing
| Item | Price |
|---|---|
| Input | $4.00 per 1 million tokens |
| Cached input | $0.20 per 1 million tokens |
| Output, including reasoning tokens | $20.00 per 1 million tokens |
| Web search | $2.50 per 1,000 queries |
| Rate limit | 3,000 requests and 4 million tokens per minute per team |
| Contributor mode (lower prices in exchange for data) | Not available for Wonder |
Source: Meta Model API pricing, model ID muse-wonder-1.0. OpenRouter (meta/muse-wonder) lists the same prices. For comparison, Muse Spark 1.3 costs $1.25/$4.25, or $0.10/$0.20 in Contributor mode.
Muse Wonder price compared with other models
| Model | Input/output, USD per 1 million tokens | Artificial Analysis index |
|---|---|---|
| GPT-6.1 Sol | 2 / 10 | 52 |
| Claude Sonnet 5.5 | 2 / 10 | 56 |
| Muse Wonder | 4 / 20 | 52 |
| Claude Opus 5.5 | 4 / 20 | 58 |
| GPT-6 Astra | 10 / 50 | 53 |
| Claude Fable 5.1 | 10 / 50 | 53 |
On paper, Muse Wonder looks like poor value: it costs the same as Opus 5.5, which scores 6 points higher on the independent index, and twice as much as Sonnet 5.5 and Sol. But the price is based on tokens, and models do not all use tokens in the same way.
What I actually paid for Muse Wonder
On October 12, I ran Muse Wonder through several tests. I used OpenRouter for all of them and took the costs from its billing records:
| Test | Muse Wonder | Claude Opus 5.5 |
|---|---|---|
| One short question (translation of the Polish test: 17% of 2,350) | $0.005 | - |
| 13 writing tasks in Polish (low reasoning) | $0.45 | $0.37 |
| 23 tasks: agent, code, Polish, reasoning, document | $0.64 | $0.75 |
| Question about a document containing 80 sets of Polish minutes | $0.19 | $0.35 |
| AI graduation exam (4 CKE papers) | $1.54 | $1.13 |
Two factors pull in opposite directions:
- A tokenizer that is efficient with Polish. The same Polish document used 46,929 tokens with Meta, 55,371 with GPT-6 Astra, and 87,675 with Claude. For long Polish input texts, Muse Wonder can cost as much as half as much as Opus, despite the same per-token price.
- A lot of reasoning. You cannot turn reasoning off (the lowest setting is minimal), and even at low, Muse Wonder reasons extensively: across 13 writing tasks, 16,157 of 21,557 output tokens were reasoning tokens. That makes it more expensive than Opus on short tasks that require long answers.
Bottom line: Muse Wonder can be a good deal for long Polish documents as input, such as contracts, reports, and meeting minutes, but less so for short questions and generating long text. The tests are covered in more detail in my test across 23 tasks and my Polish-language Muse Wonder test. We tested in Polish; these results are not an English-language benchmark.
Muse Code and Muse app subscription plans
| Plan | Monthly price | Muse Wonder Full |
|---|---|---|
| Muse Code Everyday Usage | $5 | No |
| Muse Code High Usage | $15 | No |
| Muse Code Power Usage | $50 | Yes |
| Muse Code Maximum Usage | $200 | Yes |
| Muse app (the $20 and $100 plans) | $20 / $100 | Wonder “coming soon” for subscribers; US and Canada only |
USD prices are from Meta’s developer site and CNBC reports. The Muse app was not available in Poland. For current availability and access options, check with the provider; my guide to using Muse Wonder from Poland is here.
How to spend less on Muse Wonder
- Set reasoning to low (or minimal) for simple tasks - the default is medium.
- Use caching: repeating the start of a prompt, such as the same document, costs $0.20 instead of $4 per million tokens.
- Set an output limit (max_tokens) so a verbose model does not burn through your budget. Artificial Analysis notes that Muse Wonder generates a lot of tokens compared with the median.
- Choose a cheaper model for simple tasks: in my test across 23 tasks, Gemini 3.8 Flash completed 99.5% for $0.09 - but it is not suitable for bibliographies because it invents sources.
Muse Wonder pricing: frequently asked questions
How much does Muse Wonder cost?
The API costs $4 per million input tokens and $20 per million output tokens, with cached input at $0.20 per million tokens. Wonder Full is available only with Muse Code plans costing $50 or $200 per month.
Is Muse Wonder free?
No. There is no free plan or lower-priced Contributor mode. The Muse app is free to use only in the US and Canada, and Wonder is expected to be available there first to paying subscribers.
Is Muse Wonder cheaper than Claude Opus 5.5?
The per-token price is the same. Muse Wonder uses fewer tokens on Polish text, so in my test across 23 tasks it cost 15% less ($0.64 compared with $0.75).
Sources
- Meta Model API: pricing and rate limits
- Muse Code: plans
- OpenRouter: meta/muse-wonder
- Artificial Analysis: Muse Wonder
- CNBC: Muse subscriptions (October 12, 2026)
- My measurements through OpenRouter, October 12, 2026
