PRODUCTION SCENARIO
Ostrava Legal's monthly invoice shows 3.1 million output tokens for a contract-review job that runs a reasoning model at high effort. The visible answers add up to roughly 400,000 tokens.
What explains the gap?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Reasoning tokens are billed as output tokens and reported under output_tokens_details in the usage object
Reasoning models generate hidden reasoning tokens that occupy the context window and are billed as output tokens even though they are not returned; the usage object reports their count under output_tokens_details. Tool definitions and cached prefixes are input-side costs.
Official Source: https://developers.openai.com/api/docs/guides/reasoning ↗