Reasoning models cost 30x more per query than fast inference. Learn how to route, meter thinking tokens, and price two-tier AI without billing surprises.