Oct 15, 2026
Meta Muse Wonder Pricing: API Cost, Muse Code Plans ($50, $200), and How It Compares
Cost Optimization
Distributed Inference
Muse Wonder costs $4 per million input tokens and $20 per million output on Meta's API. Muse Wonder Full is subscription only, at $50 or $200 a month in Muse Code. What each gets you, and what the same workload costs elsewhere.
Meta's first frontier models arrived on October 12 with two price tags: a per-token API for Muse Wonder, and a monthly plan for the stronger Muse Wonder Full. Here's what each costs and what you get.
Meta released Muse Wonder and Muse Wonder Full on October 12, 2026, calling them its most capable models to date and the successors to Muse Spark. The two names hide a split that matters more than the benchmarks: only Muse Wonder is on Meta's pay-as-you-go API. Muse Wonder Full, the model Meta's launch table leads with, is sold by subscription through Muse Code and nothing else.
Neither model has open weights, so there is no self-hosting path. This post covers the prices, what each access route includes, what Meta claims on benchmarks, and what the same workload costs on the open models people usually compare them against.
The short version
- Muse Wonder API: $4 per million input tokens, $0.20 per million cached input, $20 per million output. 1M-token context. Model ID muse-wonder-1.0.
- Muse Wonder Full: no API. Included in the Power Usage ($50 a month) and Maximum Usage ($200 a month) plans of Muse Code.
- Open weights: none for either model.
- Against GPT-6 Astra: Muse Wonder is 60% cheaper per token on both input and output.
- Against open models: DeepSeek V4 Pro on Yotta AI Gateway is about a quarter of Muse Wonder's input price and a seventh of its output price.
Muse Wonder API pricing
| Rate | Muse Wonder (muse-wonder-1.0) |
| Input | $4.00 per 1M tokens |
| Cached input | $0.20 per 1M tokens |
| Output | $20.00 per 1M tokens |
| Context window | 1M tokens |
| Inputs | Text, images, documents, video |
| Access | Meta Model API, keys at dev.meta.ai |
Two things to note. First, the cached-input rate is a twentieth of the standard rate, which makes a big difference for agents that resend the same system prompt and tool definitions on every turn. Second, Meta's developer post says the API is OpenAI-compatible: existing OpenAI SDK code can switch by changing the base URL and the model name. That's the same pattern Yotta AI Gateway uses, so a routing layer that already handles one can handle the other.
Meta's developer post doesn't publish rate limits, maximum output length, or a web-search price, and it lists audio input as unconfirmed. Treat those as open until Meta's docs say otherwise.
Muse Wonder Full: subscription only
Muse Wonder Full is the stronger of the two models, and Meta has chosen not to sell it per token. The only way to use it is a Muse Code subscription.
| Muse Code plan | Price per month | Muse Wonder Full |
| Everyday Usage | $5 | Not included |
| High Usage | $15 | Not included |
| Power Usage | $50 | Included |
| Maximum Usage (new) | $200 | Included, with 80 times the Everyday usage allowance |
The Everyday and High prices are from Muse Code's September plan launch, and the Power and Maximum prices are from launch coverage of Muse Wonder. Meta's own post confirms which plans include Wonder Full and the 80x figure for Maximum, but not the dollar amounts, so check the Muse Code page before you commit.
The usage allowances reset on the same five-hour and weekly cycles as the Everyday plan. None of the plans is unlimited, and Meta hasn't published the request counts behind the multiples.
If you need Wonder Full inside your own application rather than inside Muse Code, there is no supported route today. Meta hasn't said whether an API will follow.
What Meta claims on benchmarks
Meta's launch table puts both models against GPT-6 Astra and Anthropic's current models at maximum reasoning effort. These are the rows that matter most for the people reading a pricing post.
| Benchmark | Muse Wonder Full | Muse Wonder | GPT-6 Astra | Claude Opus 5.5 |
| DeepSWE 1.1 | 79.0 | 78.8 | 74.1 | 74.2 |
| Terminal-Bench 4.0 | 57.1 | 52.0 | 59.1 | 59.6 |
| Harvey LAB (legal) | 13.1 | 10.0 | 8.6 | 4.2 |
| Artificial Analysis Intelligence Index | 56 | 52 | 53 | 58 |
Read it with two caveats. The table mixes evaluations Meta ran itself with third-party numbers, and the models were run at different reasoning settings, so the rows aren't all comparable. And the numbers for the other labs' models are Meta's reporting, not those labs' own. The Intelligence Index row is the only one from an independent source, and there Wonder Full sits between GPT-6 Astra and Opus 5.5 while the API-available Muse Wonder sits one point below Astra.
The practical read: on the coding benchmark the two Muse models are nearly identical (79.0 against 78.8), and the API model is the cheaper one. On the broader index, the subscription-only Full model is the one that reaches the frontier.
What a real workload costs
Take 10 million input tokens and 2 million output tokens a day, no caching, 30 days.
| Model and route | Input / output per 1M | Per day | Per month |
| DeepSeek V4 Pro (Yotta AI Gateway) | $0.99 / $2.97 | $15.84 | $475.20 |
| GLM 5.3 (Yotta AI Gateway) | $1.40 / $4.40 | $22.80 | $684.00 |
| Qwen3.8-Max (Yotta AI Gateway) | $1.50 / $4.50 | $24.00 | $720.00 |
| Kimi K3 (Yotta AI Gateway) | $3.00 / $15.00 | $60.00 | $1,800.00 |
| Muse Wonder (Meta API) | $4.00 / $20.00 | $80.00 | $2,400.00 |
| GPT-6 Astra (OpenAI API) | $10.00 / $50.00 | $200.00 | $6,000.00 |
Muse Wonder sits between the open models and GPT-6 Astra: 40% of Astra's bill, and three to five times the bill of DeepSeek V4 Pro, GLM 5.3, or Qwen3.8-Max. With caching on a prompt-heavy agent workload the Muse Wonder number drops a lot, since cached input is $0.20, but the output rate is where the money goes and that doesn't cache.
Muse Wonder Full doesn't fit in this table. At $200 a month with a usage cap, it's a seat for a person using Muse Code, not a rate for an application. The GPT-6 Astra pricing post covers the same split on OpenAI's side, where the ChatGPT Pro plans run $100 and $200 a month.
Which one to use
Use Muse Wonder on the API if you want Meta's model in your own product and the coding-benchmark story is what you care about. It's within a point of Wonder Full there and costs 60% less than GPT-6 Astra.
Pay for Muse Code if you want Wonder Full and you're using it as a person, not as an application. $50 gets you in, $200 gets you the most usage.
Route to open models for the volume. For most traffic the capability difference between a frontier model and DeepSeek V4 Pro or GLM 5.3 won't show up in the output, and the price difference is three to five times. On Yotta AI Gateway they sit behind one OpenAI-compatible key, so testing them against Muse Wonder is a model-string change in the same client. The best OpenAI API alternatives post covers that routing pattern in detail.
Frequently asked questions
How much does Muse Wonder cost? On Meta's API, $4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens. The model ID is muse-wonder-1.0.
How much does Muse Wonder Full cost? It isn't sold per token. It's included in the Power Usage ($50 a month) and Maximum Usage ($200 a month) plans of Muse Code.
Is Muse Wonder Full available through the API? No. Meta says it is available exclusively by subscription in Muse Code, and hasn't announced an API.
Is Muse Wonder open source? No. Neither Muse Wonder nor Muse Wonder Full has open weights, so neither can be self-hosted.
What is the difference between Muse Wonder and Muse Wonder Full? Meta hasn't published a technical breakdown. In its benchmark table Wonder Full scores higher across the board, by less than a point on DeepSWE 1.1 and by four to five points on Terminal-Bench 4.0 and the Artificial Analysis Intelligence Index.
Is Muse Wonder cheaper than GPT-6 Astra? Yes, by 60% on both input and output: $4 against $10 per million input tokens, and $20 against $50 per million output.
What is Muse Wonder's context window? 1M tokens, for both Wonder and Wonder Full, per Meta.
Can I use Muse Wonder with the OpenAI SDK? Meta says yes: change the base URL to Meta's Model API and set the model to muse-wonder-1.0.
Is Muse Wonder on Yotta? The open models it competes with on price, including DeepSeek V4 Pro, GLM 5.3, Qwen3.8-Max, and Kimi K3, are live on Yotta AI Gateway with flat per-token pricing.
Bottom line
Meta priced Muse Wonder to sit under GPT-6 Astra: $4 and $20 against $10 and $50, with a cheap cached-input rate for agents. It kept the stronger Muse Wonder Full off the API entirely and put it behind $50 and $200 Muse Code plans, which makes it a product for people rather than a model for applications.
For a developer the decision is the usual one. Muse Wonder is a credible frontier option at a mid-tier price, and the open flagships on Yotta AI Gateway cost three to five times less for the traffic that doesn't need a frontier model. Run both through one OpenAI-compatible client and let cost per result decide.
