Reasoning tokens and cost

Guide · sources linked inline · 2026-10-26

Reasoning models generate internal “thinking” tokens before the visible answer. Providers bill these as output tokens, and output is typically 4–8× more expensive than input on the models we track.

How to budget

1. Run real prompts and read the output/reasoning token counts from the API usage field.
2. Enter that total (visible + reasoning) as “Output tokens per call”.
3. Where the API lets you set reasoning effort, compare low vs high on quality and cost.

Output-to-input price ratio

Next step → Model it in the calculator

Prices last checked 2026-10-26 against official sources · how we verify