# Muse Wonder: Pricing, Benchmarks, Cost per Task
> Muse Wonder costs $4 and $20 per million tokens, Claude Opus 5.5's exact rate. It scores 52 on the AA index at $4.52 a task; GPT-6.1 Sol scores 52 for $0.72.
Source: https://capitalandcompute.net/blog/muse-wonder-pricing-benchmarks/
Published: 2026-10-13
Updated: 2026-10-13
Muse Wonder costs $4 per million input tokens and $20 per million output tokens, the same rate card as Claude Opus 5.5\. On the independent Artificial Analysis index it scores 52 for $4.52 a task: about a quarter cheaper than Opus 5.5, which scores 58, and six times the $0.72 that GPT-6.1 Sol charges for the same 52\. Its case rests on long-horizon agent work, and those numbers are Meta’s own. The stronger Muse Wonder Full has no API at all.
Meta [introduced Muse Wonder and Muse Wonder Full](https://about.fb.com/news/2026/10/introducing-muse-wonder-models/) on October 12, 2026, calling them its “most capable models to date”. They follow four Muse Spark releases since April and are the first Meta models to sit in the top tier of the independent index.
| | Muse Wonder | Muse Wonder Full |
| ---------------------- | ------------------------------------------------ | --------------------------------------------- |
| Input, per Mtok | $4.00 | No per-token price |
| Output, per Mtok | $20.00 | No per-token price |
| Cached input, per Mtok | $0.20 | No per-token price |
| Where you can use it | Meta Model API (preview) and all Muse Code plans | Muse Code Power ($50) and Maximum ($200) only |
| API model ID | muse-wonder-1.0 | None |
| Context window | 1,048,576 tokens | Same |
| AA Intelligence Index | 52 (max effort) | 56 (max effort) |
| AA cost per index task | $4.52 | Not published (no public API) |
## How much does Muse Wonder cost?
One price, with no introductory discount. The [Meta Model API pricing page](https://dev.meta.ai/docs/pricing-rate-limits) lists `muse-wonder-1.0` at $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. Meta states there is “no long-context premium”, so a prompt that fills most of the 1M-token window bills at the same rate as a short one. No cache-write charge is listed. Rate limits are 3,000 requests and 4 million tokens per minute per team.
That is exactly the [Claude Opus 5.5 rate card](/blog/claude-opus-5-5-pricing-benchmarks/), and it is also the standard rate Google has set for [Gemini 4 Argon](/blog/gemini-4-argon-pricing-benchmarks/) once its introductory period ends. Three labs now sell their top API model at $4 and $20\. Against Meta’s own catalogue it is a big step up: Muse Spark 1.3 stays on sale at $1.25 and $4.25, so Wonder costs 3.2 times as much per input token and 4.7 times as much per output token.
The discount route Meta uses for Spark is not on offer here. Spark has a contributor tier at $0.10 and $0.20 in exchange for Meta training on your traffic, which [the Muse Spark 1.3 pricing breakdown](/blog/gemini-3-8-flash-muse-spark-1-3-pricing/) covers. Meta’s pricing page says Contributor mode is “not available” for Muse Wonder on the API.
## What does a task cost on Muse Wonder?
$4.52 per task on the [Artificial Analysis Intelligence Index](https://artificialanalysis.ai/models/muse-wonder), at max effort, against Meta’s list price. That puts it in the middle of the frontier field: cheaper than every Anthropic model above it, dearer than every OpenAI and Google model level with it.
__Cost per Artificial Analysis index task, frontier models, October 2026__
| Item | Value |
| ---------------------- | ----- |
| Claude Fable 5.1 (53) | $7.63 |
| Claude Opus 5.5 (58) | $5.98 |
| Claude Sonnet 5.5 (56) | $5.46 |
| Muse Wonder (52) | $4.52 |
| GPT-6 Astra (53) | $3.26 |
| Gemini 4 Argon (53) | $1.99 |
| Muse Spark 1.3 (48) | $1.60 |
| GPT-6.1 Sol (52) | $0.72 |
Cost per Artificial Analysis Intelligence Index task at each model's highest-scoring effort setting, read October 13, 2026\. The index score is in brackets. Gemini 4 Argon is costed at its introductory rate.Source: Artificial Analysis models leaderboard and model pages, read 2026-10-13
The comparison with Opus 5.5 is the cleanest one available, because the two rate cards are identical. Any gap in cost per task comes from how many tokens each model spends. On the index Opus 5.5 generated 260 million output tokens and Muse Wonder 130 million, half as many, according to the [Artificial Analysis Opus 5.5 page](https://artificialanalysis.ai/models/claude-opus-5-5). Yet Wonder’s task cost is only 24% lower, because output is not the whole bill: input and cached context are charged on every turn of a long task, and those Artificial Analysis does not break out.
Against its own predecessor the trade is steep. Muse Spark 1.3 at max effort scores 48 for $1.60 a task, so Wonder buys four index points for 2.8 times the cost. And GPT-6.1 Sol, the $2 and $10 model OpenAI shipped at [DevDay on September 29](/blog/openai-devday-2026-announcements-prices/), reaches the same 52 for $0.72 on the [Artificial Analysis leaderboard](https://artificialanalysis.ai/leaderboards/models). If the index is your measure, Wonder is not the value pick.
Speed is middling. Artificial Analysis measures 87.9 output tokens per second on Meta’s API, 84th of 185 models, with a 19.3-second wait for the first token at max effort. Opus 5.5 runs at 95.6 tokens per second.
## What is the difference between Muse Wonder and Muse Wonder Full?
Muse Wonder Full is the larger, stronger model, and you can only use it inside Meta’s Muse Code subscriptions. The [Meta models documentation](https://dev.meta.ai/docs/models) states that Muse Wonder Full is “available only in Muse Code with a subscription”, on the Power Usage and Maximum Usage plans. It is not on the API, and setting Muse Wonder’s reasoning effort to max does not select it.
The two models are separate. Artificial Analysis scores Wonder Full 56 at max effort, [fourth of 689 models](https://artificialanalysis.ai/models/muse-wonder-full), level with Claude Sonnet 5.5 and two points behind Opus 5.5\. It generated 140 million output tokens on the index, close to Wonder’s 130 million. It has no cost per task because there is no public price to compute one from.
The [Muse Code plans](https://dev.meta.ai/products/muse-code/) are priced in four steps:
| Muse Code plan | Monthly price | Usage | Muse Wonder Full |
| -------------- | ------------- | ------------ | ---------------- |
| Everyday Usage | $5 | Baseline | No |
| High Usage | $15 | 5x Everyday | No |
| Power Usage | $50 | 20x Everyday | Yes |
| Maximum Usage | $200 | 80x Everyday | Yes |
Muse Wonder itself is on every plan. One setting matters before you sign up: Meta’s documentation says Contributor mode, which gives Muse Code subscribers more usage in exchange for letting Meta train future models on their content, is on by default. It can be turned off in settings. Pay-as-you-go API traffic in Standard mode, Meta says, is “not used to train Meta models”.
## Can you use Muse Wonder through the API?
Yes, today, without a waitlist. Muse Wonder is available on the Meta Model API as a self-serve, pay-as-you-go model under the ID `muse-wonder-1.0`, labelled as a preview in Meta’s documentation. It accepts text, image, video, audio and PDF input and returns text, with reasoning effort settable from minimal to max. Meta has not published a maximum output length or a knowledge cutoff.
The Muse consumer app is next. Meta says Muse Wonder “will soon power experiences in Muse”, with subscribers first, and gives no date.
This site’s [AI model tracker](/ai-models/) carries Muse Wonder as a preview at $4 and $20, so it now appears in the [cost-per-task calculator](/ai-coding-cost-calculator/). Muse Wonder Full is listed as released with no per-token price and stays out of the calculator.
What would change this page
If Meta opens Muse Wonder Full on the API, it gets a price, a cost per task and a calculator row, and the comparison below gets re-run. If Muse Wonder leaves preview at a different rate, the cost figures above move with it.
## Is Muse Wonder better than Claude Opus 5.5?
Not on the independent index: Opus 5.5 scores 58 to Muse Wonder’s 52 at the same price per token. Meta’s case is narrower. On its [Muse Wonder model card](https://research.meta.ai/static/muse-wonder-model-card), Meta reports Wonder ahead of Opus 5.5 on long-horizon coding and on its own memory benchmark, and behind it on most other agentic tests. Every score below is vendor-reported, all at max effort except where marked.
| Benchmark (vendor-reported) | Muse Wonder Full | Muse Wonder | Claude Opus 5.5 | GPT-6 Astra |
| --------------------------- | ---------------- | ----------- | --------------- | ----------- |
| SWE-Marathon | 57.5 (xhigh) | 53.1 | 46.3 | 42.5 |
| DeepSWE 1.1 | 79.0 | 78.8 | 74.2 | 74.1 |
| MemoryBench (Meta internal) | 85.5 | 85.6 | 81.5 | 81.4 |
| Harvey LAB-AA | 13.1 | 10.0 | 4.2 | 8.6 |
| Terminal-Bench 4.0-AA | 57.1 | 52.0 | 59.6 | 59.1 |
| ProgramBench | 59.0 | 50.0 | 70.5 | 52.5 |
| OSWorld V2.1 (partial) | 78.4 | 72.6 | 81.8 | 75.8 |
| HealthBench Pro | 62.3 | 60.5 | 65.6 | 65.3 |
Two caveats sit in Meta’s own card. The 57.5 that leads most coverage is Wonder Full at xhigh effort, not max, and Meta notes the SWE-Marathon row is “not a controlled common-harness comparison”. The DeepSWE figures are not on the [DeepSWE leaderboard](https://deepswe.datacurve.ai/), which runs every model on one harness; as of October 13 that board, last updated September 22, lists neither Muse Wonder nor Opus 5.5, and its top score is a 74% three-way tie.
So the honest reading: on the long, multi-step coding Meta chose to highlight, Wonder may beat Opus 5.5, and on terminal work, program synthesis and computer use it trails it. Opus 5.5 costs $1.46 more per index task.
## How does Muse Wonder compare with GPT-6 Astra?
GPT-6 Astra is the cheaper model per task and scores one point higher. Astra charges $10 and $50 per million tokens, two and a half times Wonder’s rate, but finishes an index task for $3.26 against Wonder’s $4.52 and scores 53 to Wonder’s 52\. Astra spends far fewer tokens per task, which more than covers its higher rate. The breakdown is in [GPT-6 Astra: pricing, benchmarks, cost](/blog/gpt-6-astra-pricing-benchmarks/).
On Meta’s own table Wonder beats Astra on SWE-Marathon, DeepSWE and Meta’s memory test, and loses on Terminal-Bench, OSWorld and HealthBench. Meta also reports Wonder ahead of Astra on Vals’ Finance Agent benchmark, 56.6 to 53.5, a result [Vals’ own Finance Agent leaderboard](https://www.vals.ai/benchmarks/finance%5Fagent) does not yet show.
## Should you switch to Muse Wonder?
Only if your work looks like the work Meta benchmarked. The decision comes down to four cases:
* **Long-running coding agents, multi-hour tasks.** Worth a trial. This is where Meta’s numbers are strongest and where half the output tokens of Opus 5.5 could matter. Test on your own tasks before trusting SWE-Marathon.
* **General agentic work at the top of the index.** Stay on Opus 5.5 or Sonnet 5.5\. Both score higher, and Sonnet 5.5 costs $5.46 per index task for 56.
* **Cost-sensitive work at this capability level.** GPT-6.1 Sol reaches the same 52 for $0.72 a task, and GPT-6 Astra reaches 53 for $3.26.
* **You already pay for Muse Code.** Wonder is included on every plan, and Wonder Full from $50 a month. Check the Contributor mode setting first if your code is not yours to share.
For anything else, the fastest way to check is to price your own task shape in the [AI model comparison tool](/ai-model-comparison/), which now includes Muse Wonder against Opus 5.5.
## Frequently asked questions
How much does Muse Wonder cost?
Muse Wonder costs $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens on the Meta Model API, with no long-context premium and no introductory discount. That is the same rate card as Claude Opus 5.5.
Is Muse Wonder Full available on the API?
No. Muse Wonder Full is available only inside Muse Code, on the Power Usage plan at $50 a month and the Maximum Usage plan at $200 a month. It has no API model ID and no per-token price.
Where does Muse Wonder rank on Artificial Analysis?
Muse Wonder scores 52 on the Artificial Analysis Intelligence Index at max effort, ninth of 216 models, at $4.52 per index task. Muse Wonder Full scores 56, fourth of 689 models, level with Claude Sonnet 5.5 and behind Claude Opus 5.5 at 58.
Does Muse Code train on my code?
By default, yes, for subscribers. Meta documents Contributor mode as on by default in Muse Code: it gives more usage in exchange for Meta using your content to train future models, and it can be turned off in settings. Muse Wonder API traffic in Standard mode is not used for training.
Is Muse Wonder open source?
No. Meta has published no weights for Muse Wonder or Muse Wonder Full, and Artificial Analysis lists Muse Wonder as proprietary. Meta’s open-weight model in the Muse line is the smaller Muse Glimmer.
## Sources
Meta (2026). _Introducing Muse Wonder and Muse Wonder Full_. Meta Newsroom, October 12, 2026\. Verified 2026-10-13.
Meta (2026). _Pricing and rate limits_. Meta Model API documentation: muse-wonder-1.0 at $4, $0.20 cached and $20 per million tokens, Contributor mode not available. Verified 2026-10-13.
Meta (2026). _Models_. Meta Model API documentation: Muse Wonder preview status, context window, modalities, Muse Wonder Full availability, Contributor mode default in Muse Code. Verified 2026-10-13.
Meta (2026). _Muse Code_. Product page with plan prices and usage multiples. Verified 2026-10-13.
Meta (2026). _Muse Wonder model card_. Meta research, vendor-reported benchmark table. Verified 2026-10-13.
Artificial Analysis (2026). _Muse Wonder_ and _Muse Wonder Full_ model pages, and _Models leaderboard_. Independent benchmark index. , and Read 2026-10-13.
Artificial Analysis (2026). _Claude Opus 5.5_ model page. Read 2026-10-13.
Datacurve (2026). _DeepSWE leaderboard_. Benchmark leaderboard, v1.1, last updated September 22, 2026\. Read 2026-10-13.
Vals AI (2026). _Finance Agent benchmark_. Benchmark leaderboard. Read 2026-10-13.