Why this matters
Picture two teams running the same million input tokens through the same model. One team’s tokens are cache hits, queued through the batch API. The other team’s are fresh, sent live. On a current-generation OpenAI model, cached input runs about a tenth the price of a fresh token, and batch processing cuts whatever’s left in half.
Stack the two: at a list rate of $2 per million tokens, one team’s bill comes to 10 cents, the other’s to two dollars. It’s the same volume, but a twenty-fold gap on the invoice. (That tenth-off cache discount is current-generation only; GPT-4o’s cache discount is closer to half off, not a tenth.)
So “what does a token cost us?” turns out to be a real answerable question.
Every OpenAI cost line in CloudZero now carries the quantity behind it, and in what unit, like tokens, images, audio, or transcriptions. Divide one by the other and you have a unit price, day by day.
Usage and cost together give you two signals you can watch: your rate moving while usage sits flat (less cache reuse, or a pricier model), or usage growing while the bill doesn’t, because something you changed actually worked.
The quantity exists on every line now, so an engineering lead gets the unit price before and after a model migration, finance gets cost per unit of output for a margin model, and product gets what a single call actually costs. Same number, three answers, no console export or pivot table required.
What we built
Every OpenAI cost line now shows its own usage, in two fields:
- usage/amount is the quantity behind the dollar
- usage/units is what that quantity counts
Tokens are the common case, and that’s most of what you’ll look at. Other measures come through the way OpenAI reports them, including calls, characters, and storage. So your audio, your embeddings, and your storage charges each show up counted in whatever unit actually fits them. Nobody gets anything useful out of a token count on a storage charge.
The part that does the work is that the unit travels with the amount. When you divide, you get a price per unit you can hand to somebody without a paragraph of caveats attached to it.
How it works
The great part: there’s no configuration needed. This is on by default for every customer with an OpenAI connection.
