§
What a "credit" actually costs on a document API
Four document APIs, four different units of billing, and none of them can tell you the cost of your next call before you make it. A short tour of why every one of them ships a calculator.
Here is a question you should be able to answer about any API you depend on: what does the next call cost?
On most document APIs you cannot, and the reason is that nobody sells a request. They sell a credit, and a credit is a unit that changes shape depending on what you asked for.
The four shapes
A credit that varies by endpoint and page count. Watermarking a cover sheet and watermarking a 300-page policy are the same API call and a very different number of credits. You find out which after the fact.
A credit that varies by counting method. Some operations are counted per file, some per page, some per task, and which applies depends on the operation and sometimes on the plan.
Megabytes plus compute-seconds. Now the bill depends on how long somebody else’s servers took. A slow afternoon on their infrastructure is a line item on your invoice.
Images stored. The transform is cheap; the storage recurs. A resize you ran once in March is still billing you in November, because the output went into a bucket and the bucket has a monthly count.
Every one of these is a defensible business model, and every one of these products is decent at what it does. But all four have the same consequence: the cost of a job is knowable only after you run it, which is why all four ship a pricing calculator. A calculator on a pricing page is a confession.
The alternative is boring
One API call, $0.001. That is the whole model.
| The job | Billed as | Cost |
|---|---|---|
| Merge two 2-page invoices | 1 call | $0.001 |
| Stamp a 300-page policy | 1 call | $0.001 |
| Split a 3,000-page bundle into 40 files | 1 call | $0.001 |
| Rotate, stamp, and optimize in one pipeline | 1 call | $0.001 |
| A request we refuse for being over a limit | 0 calls | $0.000 |
There is no page multiplier, no size multiplier, no step multiplier, and no storage, because nothing is stored: the file lives in memory for the length of one request against a read-only filesystem, and then it is gone.
Where flat pricing loses
It is worth being straight about this. A competitor’s cheapest single-credit operation can beat a tenth of a cent on a one-page file. If your workload is entirely tiny documents and you have already modelled their credit table, you may come out ahead over there.
And if what you actually need is storage and a delivery CDN — the file living somewhere, served fast, for a long time — that is a different product, and we do not sell it. We process and hand the result straight back.
The rest of the time, on any real document and on any pipeline, the flat number wins, and it has the advantage of being a number you already know.