Below this text is a bill calculator. Pick a model, then say how long a typical prompt and reply are in words and how many requests you send a day. It turns your words into an estimate of tokens, multiplies by the company’s published prices, and shows a month’s bill split into its parts.

Try these, one at a time:

• Make the reply longer. Double the reply length and watch the output part of the bill. Output costs more per token.
• Make the prompt longer instead. Put the reply back and double the prompt. Each prompt token costs less, but the starting prompt is five times longer than the reply, so the two changes can add about the same.
• Turn caching up. Many apps start every prompt with the same instructions or the same document. Providers can keep that part ready and charge much less to read it again. Slide the repeated share up and watch the input part shrink.
• Switch models. Keep every setting the same and swap a small model for a large one.

Notice which part of the bill is biggest. That part is worth trimming first.