Most AI APIs are sold the way gyms are: a plan, a monthly charge, a page somewhere that lets you downgrade if you can find it, and a quiet renewal on the first of the month whether you called the API or not. Seats you did not fill. A "starter" tier that stops exactly where your project starts. A minimum commitment that made sense to somebody's spreadsheet.
We built the ParalonCloud inference API without any of that, and this post is the billing model in full, because a billing model you have to guess at is the first hidden fee.
The whole model in five lines
- You add credits, by card or crypto, from $5. A credit is a dollar.
- Every request is billed when it completes, at the model's list price, per token in and per token out, or per image.
- Credits never expire and there is nothing that renews. If you stop calling the API, you stop paying. There is no plan to cancel because there is no plan.
- Every card payment gets a numbered invoice, emailed and kept in your account.
- Before any of it, a free trial: 100,000 tokens and 5 images a day on a free key, no card required.
The rest of this post is what each line means in practice.
Billed per token, to the token
Prices are per million tokens, and they are the same numbers on the console and in this post:
| Model | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
| qwen3.8-27b | $0.12 | $1.70 |
| qwen3-3b | $0.05 | $0.20 |
| gemma3-4b | $0.04 | $0.09 |
| z-image-turbo | $0.003 per image |
A request that reads 46 tokens and writes 30 on qwen3.8-27b costs $0.00005652, and that is exactly what leaves your balance: our ledger holds eight decimals, because at these prices two decimals would round most requests to zero on one side and up on the other. Your balance in the console shows every digit that carries a value, so you can watch it move by a single request if you want to.
Nothing is charged for a request that fails. A request is refused up front,
with a 402, if your balance cannot cover it; you are never taken negative.
Settled the moment the request completes
This is the part that makes "pay as you go" true rather than a slogan. The instant the last token is written, one database transaction records the usage, debits your balance, and credits the owner of the GPU that served you. There is no end-of-month tally, no "usage will appear within 24 hours", no estimate that gets corrected later. Open Usage in the console after a request and the cost of that request is already there, per key and per model.
Your transaction history shows inference as one line per day with the request count, tokens and total, so a busy key does not produce a thousand lines of fractions of a cent. The per-request records are kept underneath; they are what the balance is computed from. We wrote up the mechanics, including why 80% of every request goes to the GPU owner, in Every Inference Request Pays the GPU That Ran It.
Add credits: card or crypto, from $5
Card. Visa, Mastercard, Amex or Apple Pay, through Stripe's hosted checkout; we never see your card number. Credits land the second the payment clears. The card fee is ours, not yours: $10 buys 10 credits.
Crypto. USDC or USDT on Ethereum or Solana, from your own wallet, with the deposit verified on-chain before it is credited. No fee from us either.
The minimum is $5 on both, low enough to try a real workload and high enough that a payment is not mostly processing fee. Credits never expire, so a $5 top-up you forget about is still there next year.
An invoice for every payment
If you buy for a company, this is usually the part that decides whether an API is usable at all. Every card payment produces an invoice: numbered in a single series, with your company name, address and VAT ID if you entered them at checkout, our company details and tax ID, emailed to you automatically together with the receipt, and available for download from your account under Transactions for as long as the account exists. Nothing to request, nothing to wait for.
What we deliberately do not have
- No subscription, no plan, no tier. Limits differ by key type (free, premium, partner) and are shown on the console before you choose; they are about rate, not about a monthly bill.
- No seats. One account, as many keys as you want, one per project if you like. Keys do not cost anything.
- No auto-renewal, no "your credits will reset". They are yours until spent.
- No minimum commitment, no annual anything.
- No "contact sales" pricing. The price per token is the price per token.
The honest part
- The $5 minimum exists because card processing has a fixed component; below that, too much of a payment would be fee. We absorb the fee either way.
- Card fees are ours; currency conversion is yours. Prices are in USD. If your card is in another currency, your bank or card network does the conversion at its own rate. That is outside our control.
- Credits are prepaid, not a line of credit. When the balance hits zero,
requests are refused until you top up. That is the point of the model, but
it means a production integration should watch the balance; the console
shows it, and the
402is explicit. - The free trial is a trial. 100,000 tokens and 5 images a day, once per account. Requests on a free key are not billed, and the GPU that served them is not paid for them either; premium traffic is what pays providers.
- We are new at card payments. The flow went live this week. If an invoice looks wrong or a payment does not credit, write to us and we will look at the actual ledger row with you.
Start
Your existing key already has its trial. When it runs out, add credits from Add Credits and create a premium key in the Console. The API, the endpoint and the models do not change; only the key does. Full billing details, including key types and limits, are in the Credits & Billing documentation.
