"Cheaper than the official API" invites an obvious question: how? This page answers it without the marketing layer, including the parts that are not flattering.
We buy model capacity in volume through upstream supply, and resell it per token at a markup we publish. We do not train models, we do not host GPUs, and we do not represent ourselves as a model vendor. We are a reseller with an OpenAI-compatible API in front.
That business only works if two things stay true: the volume price is genuinely below list, and the overhead between us and you is small. There is no per-seat layer, no subscription, no minimum commit. You prepay a balance, and tokens are billed as you spend them.
| Component | Who takes it | Roughly |
|---|---|---|
| Model cost | The upstream that serves the request | The majority of what you pay |
| Payment processing | Our payment provider | Around 4% of a top-up plus a small fixed fee |
| Withdrawal / settlement | Payment provider | 1% or a flat fee, whichever is larger |
| Platform margin | Us | The remainder, targeted at a floor we do not go below |
That last row is why "we are the cheapest on everything" is not a claim we make. There is a floor below which the arithmetic stops working, and a provider that ignores the floor is either subsidising you temporarily or not serving the model you asked for.
Because for some models our upstream price is above list — usually open-weight models served by third parties whose costs are already close to the floor. We publish those prices as they are instead of hiding them. On the pricing page you will only see the models where we are at or below list, because that page is a price list, not a catalogue. The full catalogue shows everything, including the ones where going direct is the better deal.
The upstream is not ours. That is the honest structural weakness of every reseller, and it is worth stating plainly:
model field of the response. We audit for this on our own API and publish what we find.Your balance is prepaid and is not invested, lent out, or used as working capital. It sits against capacity we have already bought. If we ever stopped operating, we would announce it in advance, keep the API answering through the notice period, and return unused balances under the refund policy. We never store your card details — payments run through our processor.
Tapix AI is operated by Tianjin Tianwei Shangxue Culture Information Consulting Co., Ltd., Room 318, No.700 Xiqing Road, Hongqiao District, Tianjin, China 300122. Support: [email protected].
See the price list · Verify a model yourself · Refund policy