An account, a key,
then credit.
Creating the account costs nothing and it issues a key in this tab. The key itself holds no requests until you fund it, and funding happens at checkout in crypto. A key with no credit returns 402. Credit is bought with crypto — $5 buys 100 requests.
Static mock. Nothing you type is transmitted, stored, or emailed anywhere, and no account is created on a server — the key is generated locally in the page.
Already holding a key? Top it up at checkout, or read the API docs first and sign up when you are ready.
The account is free. The requests are not. Fifty dollars buys thirty days of uncapped requests on this key. If you would rather look first, five dollars buys a hundred requests that never expire.
- Uncapped requests for $50 a month on this key
- Balance never expires — spend it next year if you like
- Try it for $5: a hundred requests, prepaid
- Key issued instantly, in the browser, before you pay
- One account per person
- Same model and 262,144-token context on both products
The key is generated in this tab.
The form collapses and the string appears in panel 02, on this page. Nothing is posted, nothing is emailed, and the address you typed is not stored anywhere — this build has no backend to store it in.
It arrives with an empty balance.
A fresh key holds zero requests, because no credit is issued at signup and none is ever given away. Authentication succeeds; billing does not, and the endpoint answers with an out-of-credit response rather than serving the call.
Credit lands when the chain confirms.
Send $50 for thirty days without a counter, or $5 for a hundred requests that never expire. Both are prepaid in crypto on the checkout page, and both attach to the key you are already holding.
Three routes. That is the whole API.
There is no catalogue to browse and no model picker to choose from. One model, one endpoint family, one key.
POST /v1/chat/completions
Openai-compatible body, model string qwen-flash-next, reasoning at off, low or xhigh. Streamed or buffered. One call is one request, whatever its length.
GET /v1/models
Returns exactly one entry. If your client expects a list to pick from, it will get a list of one and should stop asking for options.
GET /v1/usage
Requests and tokens against your key, plus the balance left on it. Per-key counters only — no prompt, no completion, no timestamp per call.
Send it like any other OpenAI client.
curl https://api.rawinfer.com/v1/chat/completions \
-H "Authorization: Bearer $RAWINFER_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen-flash-next",
"reasoning_effort": "off",
"messages": [{"role": "user", "content": "your prompt"}]
}'The key is the only credential. There is no organisation header, no project scope, no admin token to lose.
What a hundred requests is actually worth.
Enough to finish an integration and find the edges of it, with room left over for a real workload: the long system prompt, the tool schemas, the transcript you actually keep. A hundred calls to this endpoint at the lengths it is built for is a working day of agent traffic, not a taste of one.
When the hundred are spent the key stops answering and you buy another pack. There is no overage, no soft limit that quietly bills, and no prompt that turns into a charge without a transfer behind it.
One base URL, one model string.
The endpoint is OpenAI-compatible, so the client you already have keeps working. Point it here, paste the key, change nothing else.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.rawinfer.com/v1",
api_key=os.environ["RAWINFER_KEY"],
)
resp = client.chat.completions.create(
model="qwen-flash-next",
reasoning_effort="xhigh", # off | low | xhigh
messages=[{"role": "user", "content": prompt}],
max_tokens=8192,
)
print(resp.choices[0].message.content)Thinking is on by default. Set reasoning_effort to off when you want the shortest path to an answer and xhigh when the problem deserves the tokens.
Four things that trip people up.
- The base URL ends in
/v1. Your client appends/chat/completionsitself. Adding it twice is the most common 404 on this endpoint. - The model string is
qwen-flash-next. Exactly that. There is no dated variant, no preview alias and no other ID that resolves —/v1/modelsreturns a list of one. - Images and video frames go in as input. They are content parts in the same message array. Text comes out; there is no generation route for pictures.
- Five calls in flight. Beyond that your own requests queue behind each other. Batch loops that fan out to fifty workers will look slow until you cap the pool at five.
Full request and response schemas, streaming, and the error table are on the API docs page.
One price. Everything on the shelf.
Fifty dollars, thirty days, no counter. Paid in crypto, and it does not renew behind your back. If you want a look before a month, five dollars buys a hundred requests.
$50 / month, uncapped
One key, thirty days, no request counter. Concurrency 5, 262,144 tokens of context, every effort level. Cancel by not paying.
- Uncapped requests, fair use
- Runs thirty days from confirmation
- Same model and context window as the pack
If you are unsure, spend five dollars first and watch a week of real usage. That is what the hundred-request pack is for.
It is a bearer token. Treat it like cash.
There is no second factor, no IP allowlist and no per-device scope — the string is the account. Anyone with a copy can spend the whole balance from anywhere, and there is no dispute process for requests a stranger made with your key.
Read it from an environment variable, keep it out of the repository, and rotate it after sharing a screen. If it leaks, assume the balance is gone: the key can be replaced, the requests cannot be un-spent.
What we actually store- Environment variable, never a string in source.
- Never in a public repository, never in a screenshot.
- One key per person — sharing breaks fair use on both cards.
- No key in a browser bundle, a mobile app, or anything a user can read.
- No key in a support ticket. We never ask for one.
Small print, at readable size.
Six facts and no fine pen. None of them change after you sign up.
| Term | Value | What it means in practice |
|---|---|---|
| Credit at signup | none | Account creation is free and issues the key. Requests are bought separately, and no free credit is ever issued. |
| First price | $5 / 100 | One hundred calls to /v1/chat/completions. Length does not change the count. |
| Expiry on paid credit | never | A pack bought today spends next year. The monthly product runs thirty days and does not auto-renew. |
| Accounts per person | 1 | Serial signups on one address are merged and the duplicate keys are disabled without notice. |
| Concurrency | 5 | Five requests in flight per key, the sixth waits. Sequential clients never notice the limit. |
| Refunds | none, ever | Crypto in, crypto never out. Not for an unused pack, not for unused days of a month. |
The binding text lives in Terms. The retention promise lives in Privacy & ZDR.
Do not sign up if any of this is a requirement.
Other services will give you every item on this list. This one will not, and it is cheaper to read it before creating an account than to discover it at the payment step.
See what you do get- A card, a bank transfer, or a processor sitting in the middle of the payment.
- An invoice, a purchase order, a VAT number or a billing address.
- A refund, a free allotment, or credit back for hours the endpoint was down.
- A service-level agreement, a named legal entity to contract with, or a jurisdiction.
- Moderation, prompt review, or an audit trail of what was generated.
- A second model, a model picker, or a tier above uncapped.
One model
262,144 context
Zero retention
5 in flight
Crypto only
No refunds
Six questions we get first.
Do I need a wallet to create an account?
No. An email and a password is the whole form, and the key appears before any money is involved. A wallet only appears when you fund the key, and it appears on the checkout page, not here.
Card payments are not accepted anywhere in this flow, on any page, for either product. If you want to pay with a card, this is not the service for you.
What does an unfunded key do?
It authenticates and it answers with an out-of-credit response, because the balance it spends against is zero. Nothing is charged, no overage accrues, and no invoice appears later. The account and the key string stay valid so that a top-up lands on the same key.
Fifty dollars buys a month of uncapped traffic. Five buys a hundred requests. Both are prepaid in crypto.
Is there a free tier?
No. There is no free tier, no signup credit, and no promotional allotment of any size. The $5 pack is the cheapest thing on the price list and it is how anybody decides whether this endpoint is worth paying for.
That is a deliberate choice rather than a gap: a free allotment has to be defended against abuse, and defending it would mean watching keys the way every other provider watches them.
What do you keep about me?
An email, a password hash and the per-key counters: requests, tokens, timestamps. Prompts and completions are never written to disk and never logged, so there is no transcript to hand over even if somebody asked.
This build stores nothing at all — the form is a mock and the key is generated in the page. The privacy page describes the production behaviour in full.
One account per person — how strict is that?
Strict. Duplicate accounts on the same address are merged on key and payment history, and the extra keys stop working without notice. There is no appeals queue and no polite email about it.
If more than one person needs access, the answer is a second key on a paid product with more concurrency, not a second signup.
Do you send a confirmation email?
No. The key appears on screen the moment you submit, and that screen is the only copy you get. Copy it somewhere durable before you close the tab.
In this build the string is not written to storage at all, so a closed tab means a lost key and a fresh signup. In production the key can be reissued to the email on the account, which is the only reason the email field exists.
Still deciding? The pricing page has the metered arithmetic, and support answers anything the docs do not.
The refusals were cut
at the weight level.
Not prompted around, not fine-tuned away. Create the key now, put five dollars behind it when you have a first task, and find out what that means for the work you actually do.
No logs. No lock-in. No jurisdiction.