Lesson 6 of 7
Why thinking costs more
5 min read
A 'thinking' model gives a better answer to a hard question — and a surprisingly bigger bill. Where did the extra cost come from?
Invisible tokens you still pay for
A reasoning model works through a problem step by step before it answers. Those steps are real output — [reasoning tokens](glossary://reasoning-tokens) — generated one by one, billed at the output rate, even though the interface usually hides them. The visible answer might be short, but you paid for all the thinking behind it.
Hidden reasoning tokens are billed at the output rate — thinking isn't free.
Match the model to the question
This is worth it for genuinely hard problems — multi-step math, tricky code, careful analysis — where the better answer pays for itself. It's wasteful for easy questions a plain model would nail. The skill is knowing when a problem deserves the extra thinking, and reaching for the cheaper model the rest of the time.
Reasoning earns its cost on hard problems — and wastes it on easy ones.
Don't default every request to a reasoning model — the hidden thinking tokens can multiply the bill for questions that never needed them.
Paying for thought
- —Reasoning models emit hidden thinking tokens.
- —Those tokens are billed at the output rate.
- —Use reasoning for hard problems, cheaper models for easy ones.
Why does a reasoning model cost more than a plain one?
Continue in the app
Take the whole What AI Costs course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.