There is no free way to use Muse Wonder as of its October 12, 2026 launch. The Meta Model API bills per token with no free credits, the Muse consumer app is free but is rolling Muse Wonder out to its paying subscribers first, and Muse Code has no free plan. The cheapest routes are a few cents of API usage per test, the $5-a-month Muse Code Everyday plan, or a third-party gateway that already carries the model at Meta’s prices. Muse Wonder Full, the larger model, starts at $50 a month and has no route cheaper than that.
This post lists every route we could verify, what each one costs, what it gives you, and the traps that make “free” claims about this model wrong. If you only want to try the model, the API path at the bottom costs less than a coffee. Background on the model is in the Muse Wonder explainer.
Every route, priced
| Route | Cost | What you get | Muse Wonder Full? |
|---|---|---|---|
| Meta Model API | $4 / $20 per 1M tokens, pay as you go | muse-wonder-1.0, all three endpoints, 1M context |
No |
| Muse Code Everyday | $5 / month | Muse Wonder in the terminal agent, 10 to 50 prompts per 5 hours | No |
| Muse Code High | $15 / month | 5x Everyday usage | No |
| Muse Code Power | $50 / month | 20x usage, Muse Wonder Full | Yes |
| Muse Code Maximum | $200 / month | 80x usage, Muse Wonder Full | Yes |
| Muse app free tier | $0 | The app, without Muse Wonder for now | No |
| Muse app subscriptions | $20 or $100 / month | Muse Wonder, rolling out to subscribers first | No |
| Vercel AI Gateway | $4 / $20 per 1M, billed by Vercel | meta/muse-wonder-1.0 |
No |
| Cloudflare Workers AI | $4 / $20 per 1M, billed by Cloudflare | meta/muse-wonder-1.0 via env.AI.run |
No |
Nothing in that table is $0 with the model included. Meta’s quickstart, pricing page and developer blog mention no trial credits. The one promotion we found is the Meta Global AI Developer Hackathon, which offers $1M in prizes, not API credit.
Why there is no free tier
Meta’s launch coverage framed Muse Wonder as a test of whether users will pay. The consumer app’s free tier keeps the older Muse Spark models; Wonder goes to the $20 and $100 subscribers, and Meta has not said when or whether free users follow. Zuckerberg’s line is that Meta will bring this level of intelligence to everyone over time, without a date.
On the developer side, the Model API has always been pay-as-you-go with no sign-up credit, and Muse Code’s cheapest plan is $5. That is unusual next to Google, which keeps a free Gemini API tier, and closer to Anthropic, where Opus 5.5 also has no free route. If you see a site offering free Muse Wonder access, it is either reselling API calls at a loss, serving a different model under the name, or asking for your key.
Route 1: the API, for cents
The honest cheapest option for a developer is to pay for exactly the tokens you use. A realistic test prompt of 2,000 input tokens and 1,000 output tokens, reasoning included, costs about $0.028. Ten of them cost less than 30 cents. There is no monthly minimum and no commitment.
curl https://api.meta.ai/v1/responses \
-H "Authorization: Bearer $MODEL_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"muse-wonder-1.0","input":"Explain idempotency keys in two paragraphs.","reasoning":{"effort":"low"},"max_output_tokens":800}'
Keep effort at low for exploratory calls. Reasoning tokens bill at the $20 output rate, and at max they can be most of the response. The API guide covers the endpoints and parameters.
Two cost controls matter even at test scale. Set max_output_tokens on every request so a runaway reasoning chain cannot bill you for a 100K-token answer. And put the key in an Apidog secret variable rather than a shell history or a shared doc; a leaked key on a pay-as-you-go account is the fastest way to turn a cheap test into an expensive one.
Route 2: Muse Code Everyday at $5
If you would rather have a fixed monthly cost, the $5 Muse Code plan includes Muse Wonder along with Muse Spark and Muse Image, with 10 to 50 prompts every five hours. Install the agent with one command and pick the model:
curl -fsSL https://dev.meta.ai/install.sh | sh
muse --model muse-wonder-1.0
Know what you are agreeing to. Muse Code subscriptions default to Contributor mode, which lets Meta use your prompts, code and outputs to train future models in exchange for more usage. You can turn it off in settings, and future content stops being used. For a hobby project that trade may be fine; for client code it is not, and you should flip the switch before the first session.
The $5 plan does not include Muse Wonder Full. That model is only on Power ($50) and Maximum ($200), and setting --reasoning-effort max on Wonder does not change which model you are talking to.
Route 3: gateways you already pay
If your stack already runs through Vercel AI Gateway or Cloudflare Workers AI, both added meta/muse-wonder-1.0 at launch at the same $4 / $20 rates. That is not free either, but it means no new account, no new key, and the model shows up in the provider dropdown you already use. Vercel’s listing confirms tool calling, structured output and a 1M context; Cloudflare serves it through the Responses format with reasoning tokens reported in usage.
Route 4: wait for the consumer app rollout
The Muse app is free. Muse Wonder inside it is not, yet. Meta says the model will power app experiences soon, starting with subscribers at $20 and $100 a month. If you want to chat with Wonder rather than call it, the app subscription is the path, and the free tier may follow later. There is nothing to do today except check the app’s plan page.
Traps to avoid
- “Free Muse Wonder API” sites. No authorized free tier exists. Anything offering one is reselling, substituting, or harvesting keys.
- Confusing the three Muse products. The Muse app, Muse Code and the Model API are separate logins with separate billing. An app subscription does not give you API access, and an API key does not unlock the app.
- Contributor mode by default. Cheaper Muse Code usage comes with training on your data unless you opt out. API traffic in Standard mode is not used for training.
- Thinking you have Full. Headlines about Meta’s frontier benchmark wins describe Muse Wonder Full. The $5 plan, the API and the gateways all serve the smaller Wonder. The Full vs Wonder comparison spells out the four-point gap.
- Unbounded reasoning. Without
max_output_tokens, a hard prompt at high effort can produce tens of thousands of billable reasoning tokens on a single call.
Test it for cents in Apidog
The cheapest honest trial is also the most useful one. In Apidog, create an environment with base_url set to https://api.meta.ai/v1 and MODEL_API_KEY as a secret, add one Responses request, and duplicate it for the two or three prompts that matter to your product. Add an assertion on $.usage.output_tokens so no test can exceed a budget you set, and a post-processor that logs the dollar cost of each run. Ten real prompts at low effort will cost under a dollar and tell you more than any benchmark. Download Apidog to run them.
Frequently asked questions
Does Meta give new developers free credits? Not that we could find. The quickstart, pricing page and launch blog mention none.
Is Muse Wonder in the free Muse app? Not at launch. Meta is rolling it out to paying subscribers first and has given no date for free users.
Is there a student or startup program? None announced for the Model API. The hackathon offers prizes, not credits.
What is the cheapest way to use Muse Wonder Full? Muse Code Power at $50 a month. There is no API, no per-token price, and no cheaper plan that includes it.
Next step
Spend a dollar on the API before you spend $5 or $50 on a plan. Ten of your own prompts at low effort will show you whether the model fits your work, and the usage block will show you what it would cost at scale.
