How a request is priced

Tokens, cache, the balance hold and what you see in your profile.

Prices are in dollars per 1 million tokens, separately for input, output and cache. Input is everything you send to the model; output is what it generates.

Cache is cheaper than regular input: when a new request starts with the same context, part of the tokens is billed at the cache price. Code editors often resend the same context again and again, so cache noticeably lowers the bill.

Before a request, a hold is placed on your balance for the largest possible answer. Afterwards only the real cost is charged and the rest of the hold is released. That is why a request can return 402 even if a short answer would have fit: top up a bit more.

Your request count and spend are visible in the bot profile.