Token Pricing

A token is roughly four characters or three-quarters of a word in English. Here is the quick reference table, how to estimate from your own text in seconds, and the five cases where the ratio breaks badly.

Input tokens, output tokens, cached tokens and cache writes — what each one costs on GPT-6 Astra, what actually counts toward your input bill, and three worked examples showing where the money really goes.

Cached input tokens bill at a tenth of the standard rate. Here is exactly what gets cached, the 1,024-token minimum, why writes cost 1.25x, and the prompt-ordering mistakes that silently destroy your hit rate.