How to Read a Gateway Price List
Official base rate, actual rate, and the multiplier column — how to read a published AI gateway price list, plus the three things a good one always shows.
Most AI gateways do not publish prices. That is not an accident: opacity is a pricing feature. When a list is published, most readers still misread it — they compare the wrong column, ignore the line that dominates their bill, and never check the one thing that makes a price verifiable. This is a short field guide to reading one properly.
Disclosure up front: I operate Teapot Router, whose price book is public without login at teapotrouter.com/en/pricing. The reading method below applies to any gateway that publishes a list — including our competitors'.
The two-column rule
A publishable price list shows, for every model, the official base rate next to what a standard account actually pays — per million tokens, input and output separately. If a list shows only one column, you cannot tell whether you are looking at the official rate or the discounted one; if it shows only a multiplier, you cannot check the arithmetic at all. Two columns, per token, per model: that is the entry fee of honesty.
The line item readers skip: cache reads
Agent workloads are dominated by cache reads — the re-sent, stable prompt context billed at a steep discount. A price list that omits cache-read pricing is omitting the majority of an agent bill. When you compare two gateways, the cache-read column moves the effective rate more than the input column does, and a gateway that hides it is hiding where your money actually goes.
Three things a good list always shows
Beyond the columns, three properties separate a price list from a marketing page:
- Every public model, not a highlight reel — if the model you run is missing from the list, the list is not a price list.
- Machine-readable access — a JSON endpoint without login, so scripts (and calculators) can compute cost before you commit, not after.
- No expiry games — prices tied to a published book you can re-check anytime, not screenshots in a chat thread.
Red flags worth walking away from
Watch for three patterns. A multiplier with no base rate anywhere: unauditable by design. Ranges instead of prices — "up to 70% off" is a mood, not a number. And a list that requires a top-up before you can see it: at that point the gateway is asking you to pay for the privilege of learning its prices. None of these are technical limitations; they are choices, and they tell you what the operator expects to happen after you discover the real number.
A price list exists so you can check the invoice against it. Demand two columns per model, cache-read pricing, machine-readable access, and no login wall — anything less is a brochure.
FAQ
Where can I see Teapot Router's price list?
Public, no login: teapotrouter.com/en/pricing — every public model with the official base rate next to the actual rate, USD per 1M tokens, cache reads included.
Why do multipliers mislead?
A multiplier only means something against a base rate you can verify. Without the base column you are trusting the operator's arithmetic — and operators who skip the base usually skip it for a reason.
What is a cache-read price?
The rate for re-reading stable prompt context, usually a small fraction of the input price. On agent workloads it is often the largest line item on the bill — which is why its absence from a price list is the loudest silence in the document.