AI Token Cost Calculator

Token pricing is a straight multiplication: tokens divided by a million, times the price per million, input and output priced separately. A session of 50000 input and 15000 output tokens at 5 and 25 per million costs 0.625 — twenty-two of those a month come to 13.75.

Language models are billed by the token, and the arithmetic is simpler than the pricing pages make it look. Divide the tokens by a million, multiply by the price per million, and do it twice — once for what you send and once for what comes back, because output is charged at a higher rate than input. A session that reads 50000 tokens of context and writes 15000 tokens of code costs 0.25 for the input and 0.375 for the output: 0.625 in total. Run twenty-two of those in a working month and the bill is 13.75. This page deliberately has no price list. Provider rates move, and a page with a rate baked into it goes wrong quietly — it keeps showing a confident number long after the number stopped being true. So the price per million is a field you fill in from whatever your provider publishes today, and the page does the multiplication. That also makes it work for any provider and any model, including ones that did not exist when this page was written. Two things the figure leaves out. Caching changes the input price — a cache read is typically billed at a fraction of the base rate, a cache write at a premium — so a heavily cached agent costs less than this suggests. And server-side extras such as web searches or container time are billed separately from tokens altogether.