Frequently asked questions
Why does a long conversation cost more than its tokens?
Because the model is stateless: every turn you resend the entire prior conversation as input. By turn eight you are paying for the first turn eight times over. Total billed tokens grow roughly with the square of the turn count, so length, not just message size, drives the cost.
How do I lower cost per conversation?
Trim or summarise old turns instead of resending everything, use prompt caching so the repeated prefix is billed at a discount, cap conversation length, and route simple turns to a cheaper model. These attack the re-sent context that makes long chats expensive.