How our pricing works, and where the capacity comes from

"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.

The short version

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.

What the price is made of

ComponentWho takes itRoughly
Model costThe upstream that serves the requestThe majority of what you pay
Payment processingOur payment providerAround 4% of a top-up plus a small fixed fee
Withdrawal / settlementPayment provider1% or a flat fee, whichever is larger
Platform marginUsThe 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.

Why some models here cost more than the official list

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.

If a competitor is far cheaper on the same frontier model, one of three things is true: they found cheaper volume than we did, they are burning money to buy market share, or the answer is not coming from that model. The three checks here tell you which.

What we do not control, and how we handle it

The upstream is not ours. That is the honest structural weakness of every reseller, and it is worth stating plainly:

What happens to your prepaid balance

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.

Company information

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

Last updated: 2026-11-20最后更新:2026-11-20