Fast is built for everyday actions — short replies, quick summaries, and simple plan generation. It starts at 1 credit. Thinking is the deeper-reasoning model: better for complex multi-step plans, long lookback windows, and substantive clarifying questions. It starts at 5 credits, because it reads and reasons over far more before answering. Both rise with how much data the request carries and how long the answer is, so a quarter costs more than a week on either model. You can switch between them on every chat input via the model pill, so you only pay the higher rate when you actually want the deeper response — and before anything runs you see an estimate, and you're charged for what the AI actually uses. With Claude, GPT and Grok in Thinking, an answer cut short after it has started (by Stop, by closing the page, or by a lost connection) is charged the most it could cost, because their thinking cannot be counted once it is cut. A plan or table starts only when your credits cover the most it could cost, and one cut short is charged at most its starting price.