Meta AI models · Developer cost guide
Muse Wonder API Pricing: Cost Calculator and Muse Spark Comparison
Muse Wonder is Meta’s newer model for long-running agents, coding and complex multi-step work. Its published API price is $4 per million uncached input tokens and $20 per million output tokens, with cached input at $0.20 per million. Use the calculator below to estimate your token bill and compare it with Muse Spark 1.3 before choosing a model. Source: Meta’s official Muse Wonder model page.
Muse Wonder API cost calculator
Enter the tokens and web-search grounding queries you expect in a month. Put cached input in its own field; do not include those same tokens again in uncached input. The estimate uses Meta’s published standard API rates and compares the same workload with standard Muse Spark 1.3 pricing.
Estimate your monthly token cost
This estimates token charges plus Meta’s published web-search grounding charge when you enter a query count. It excludes taxes, retries, other tools, third-party platform fees, orchestration overhead and any other charges that Meta may add or change. Actual agent workloads can generate more input/output tokens than expected. Muse Wonder Full is not included because Meta lists it as a Muse Code subscription model without a public per-token rate.
Official Muse Wonder API rates
| Model / rate | Price per 1 million tokens | What it means |
|---|---|---|
| Muse Wonder input | $4.00 | Input tokens that are not billed at the cached-input rate. |
| Muse Wonder cached input | $0.20 | Input tokens eligible for the cached-input rate. |
| Muse Wonder output | $20.00 | Tokens generated by the model. |
| Muse Spark 1.3 standard input | $1.25 | Standard mode listed by Meta as not used to improve its products. |
| Muse Spark 1.3 standard cached input | $0.15 | Standard mode cached-input rate. |
| Muse Spark 1.3 standard output | $4.25 | Standard mode output rate. |
Rates are the displayed USD prices on Meta’s model pages at the last review date. Meta also lists a separate lower-priced Muse Spark 1.3 Contributor mode—$0.10 input, $0.002 cached input and $0.20 output per million tokens. Meta says Contributor mode permits the prompts, code, files, context and responses you send to be used to train future Meta models. It is intentionally not used as the calculator’s default comparison because it has a materially different data-use condition. Muse Wonder has no Contributor mode on the Model API. Review the current official pricing and rate-limit documentation, Muse Wonder rates and Muse Spark 1.3 rates before making a production decision.
Web search grounding and API rate limits
Token usage is not always the full bill. Meta lists web-search grounding at $2.50 per 1,000 search queries for text models including Muse Wonder and Muse Spark. That is charged in addition to the request’s token cost. Enter your monthly search-query count in the calculator to include this line item in both model estimates. A model call that does not use web-search grounding should not be assigned this search charge.
Meta’s published rate-limit table lists muse-wonder-1.0 at 3,000 requests per minute and 4,000,000 tokens per minute per team. Meta says requests, tokens and streams across a team’s API keys count toward the team’s shared limits; each model has its own limit. Your account and live documentation are authoritative if these figures change. See Meta’s pricing and rate-limit documentation.
Meta also states that token-priced models do not have a separate long-context premium: the rate remains the same whether the context window is mostly empty or nearly full. Billing still follows the actual input and output tokens, with eligible cached-prefix input billed at the lower cached rate. Check the response’s cached_tokens value rather than assuming a prompt was cached.
How much does Muse Wonder cost for common token volumes?
These examples assume no cached input, no tool charges and no additional fees. They show the API token bill only, not the total cost of a completed agent task.
| Uncached input | Output | Muse Wonder | Muse Spark 1.3 standard |
|---|---|---|---|
| 100,000 | 10,000 | $0.60 | $0.17 |
| 1,000,000 | 100,000 | $6.00 | $1.68 |
| 10,000,000 | 1,000,000 | $60.00 | $16.75 |
Output matters: at the listed standard rates, one million output tokens cost $20 with Muse Wonder versus $4.25 with Muse Spark 1.3. Cached inputs may reduce a bill, but only tokens Meta counts as cached qualify for that rate. A million-token context window is a maximum context capacity, not a monthly allowance and not a promise that each request will use or cost one million tokens.
Muse Wonder vs Muse Spark 1.3: which should you choose?
| Decision factor | Muse Wonder | Muse Spark 1.3 |
|---|---|---|
| Best starting point | Complex, long-running agent work, advanced coding and computer-use workflows where the additional capability is worth testing. | Cost-sensitive workloads, existing agent flows and tasks where its results already meet your quality bar. |
| Context window listed by Meta | 1 million tokens | 1 million tokens |
| Standard input per 1M tokens | $4.00 | $1.25 |
| Standard cached input per 1M tokens | $0.20 | $0.15 |
| Standard output per 1M tokens | $20.00 | $4.25 |
| Model identifier | muse-wonder-1.0 | muse-spark-1.3 |
| Availability | Meta Model API and Muse Code | Meta Model API and Muse Code |
At identical standard token volumes, Muse Wonder costs 3.2 times as much for uncached input and about 4.7 times as much for output. That does not automatically make it worse value. If Wonder completes a difficult task in fewer attempts, needs less human correction or succeeds where Spark fails, total cost per successful result could look different. Measure both models on your own representative prompts before committing.
What do independent benchmarks show?
Artificial Analysis reported Intelligence Index v4.3 scores of 52 for Muse Wonder and 56 for Muse Wonder Full at maximum reasoning effort, compared with 48 for Muse Spark 1.3 in that evaluation. The same report found higher task costs for Muse Wonder on its own index workload than for Muse Spark 1.3. These scores are useful external context, not a guarantee of performance on your data or codebase. Read the independent benchmark and cost analysis and check the methodology before using scores to choose a model.
Muse Wonder Full price: API or subscription?
Muse Wonder Full is distinct from the standard Muse Wonder API model. Meta lists Wonder Full as available through Muse Code subscription plans and does not show a per-token API rate for it on the model page. At the time checked, the listed Power Usage plan is $50 per month and includes Wonder Full with 20× the Everyday Usage limit; Maximum Usage is $200 per month and lists 80× the Everyday Usage limit. Plan inclusions, quotas and prices can change, so confirm them on Meta’s current model and plan page.
Do not compare a subscription’s monthly fee directly with an API bill without estimating your actual volume and considering the subscription’s usage limits. A subscription is not unlimited usage, and Meta does not publish Wonder Full as a per-token price in the cited model table.
How to access Muse Wonder
- For API development: start at the official Muse Wonder model page, select the Meta Model API route and follow Meta’s quick-start and documentation links.
- For terminal-based coding: review Meta’s Muse Code setup guide. Wonder Full access is tied to eligible subscription plans shown by Meta.
- For an end-user assistant: do not assume the consumer Muse app is billed at API token rates. Meta announced that Wonder would be rolled into Muse experiences starting with subscribers; check the app and announcement for current rollout status.
- Before launch: set usage alerts or budget controls where available, log token counts, test a representative workload, and evaluate the completed task—not just the price per million tokens.
A better way to estimate real agent costs
For agent workflows, the prompt itself is only part of the expense. An agent may read a large repository, call a tool, inspect tool output, retry after an error and generate a final answer. Each model call can consume additional input or output tokens. To build a practical forecast:
- Run a small sample of real tasks and save the token usage for every call.
- Separate cached from uncached input using the billing definitions in Meta’s current API documentation.
- Include failed attempts, retries and intermediate reasoning/output in your observed total rather than estimating only the final response.
- Compare the cost per completed task and human review time for Spark and Wonder, not just their listed rates.
- Repeat the test after model, prompt, tool or pricing changes.
Muse Wonder API pricing FAQs
How much does Muse Wonder API cost?
Meta lists $4 per million uncached input tokens, $0.20 per million cached input tokens and $20 per million output tokens. The amount you pay depends on actual token usage and any additional applicable charges. Check the official rate card.
Is Muse Wonder free to use?
The API has published token-based rates; do not assume API use is free. Muse Wonder is also offered through Muse Code, which has subscription plans and separate usage limits. Check the current Meta account and pricing screens for any trial credits or changes.
How much does one million Muse Wonder input tokens cost?
One million uncached input tokens cost $4 at the published standard rate. That figure excludes output tokens, cached input, tools and any other charges.
Why is Muse Wonder more expensive than Muse Spark 1.3?
The listed standard token prices are higher for Wonder, particularly output. It may still be worth testing for complex tasks if it raises success rates or reduces retries and human correction. Compare cost per successful result on the same workload before deciding.
Is Muse Wonder Full available through the API?
Meta’s current model page lists Muse Wonder Full as available by subscription in Muse Code and does not list a public per-token API rate for it. Confirm the latest access options on Meta’s official page before implementation.
Does a one-million-token context window mean each request costs one million tokens?
No. Context window describes the maximum context a model can handle, not a minimum billed amount or monthly allowance. Charges depend on the tokens actually processed and generated under the applicable rate card.
Is Muse Wonder automatically better than Muse Spark?
No single benchmark settles that question for every task. Independent testing reports stronger overall index scores for the new Wonder models, but your own data, tools, latency requirements and budget determine which model offers better value for a specific workload.
Sources and verification
The price table and access statements above are based on primary Meta sources. The benchmark section uses independent analysis and is labelled separately from Meta’s own claims.
- Meta Model API documentation: standard and Contributor pricing, cached-input rules, web-search charges and rate limits
- Meta developer page: Muse Wonder model, context window, API rates and Muse Code plans
- Meta developer page: Muse Spark 1.3 rates and Contributor-mode data-use note
- Meta developer announcement: building with Muse Wonder and Muse Wonder Full
- Meta Newsroom: Introducing Muse Wonder and Muse Wonder Full, October 12, 2026
- Artificial Analysis: independent benchmarks, token pricing and task-cost analysis, October 12, 2026
Editorial note: UltimateInfoGuide is an independent publisher and is not affiliated with Meta. API pricing and product access can change. This guide was checked against the sources linked above on October 13, 2026; verify current rates before budgeting or deploying a production system.
For the consumer-facing assistant—its tasks, setup, plan limits and practical use—see our Meta Muse AI tasks and limitations guide.
Advertisements