Input vs output

Input and output are both tokens — so why does the reply cost several times more than the prompt you sent?

Reading is cheap, writing is dear

Input tokens — your prompt and any history — are read in one pass, so they're cheap. Output tokens — the reply — are generated one at a time, each step running the whole model again, so they typically cost around 5× more per token. The longer the answer you ask for, the more that side of the bill grows.

You re-pay the history every turn

A model has no memory of the chat between turns, so every message re-sends the whole conversation as input. Turn ten pays to re-read turns one through nine. That's why long chats get more expensive with each reply — and why starting a fresh chat for a new topic saves money.

Where the money goes

• —Input (prompt + history) is cheap; read once.
• —Output (the reply) costs ~5× — made token by token.
• —Each turn re-pays the whole history — long chats cost more.