AI calculators

Why you need to work this out

Using a model's API is not billed per request, it is billed per token, the unit these models chop text into. That makes the cost very hard to intuit: two requests that look identical can cost ten times as much depending on the context you send and the length of the reply. These calculators exist so you can estimate the bill before building on top of it, not after receiving it.

What surprises people about the bill

• Output costs considerably more than input, typically three to five times. If you generate long texts, that is where most of your spend goes.
• In a conversation, every message resends the whole prior history as input. A thirty-turn chat does not cost thirty requests: it costs a good deal more, because the last one resends the entire conversation.
• Spanish uses more tokens than English to say the same thing, roughly 20-30 % more. A service with the same volume costs more to run in Spanish.
• Small models cost a fraction of the large ones and for classifying, extracting data or summarising they give an equivalent result. Trying the cheap one first is the optimisation that saves most.