Report #98580
[cost\_intel] Reasoning tokens are invisible but billed as output and can dominate cost
Set reasoning/thinking effort explicitly, cap max\_output\_tokens, route simple queries to non-reasoning models, and inspect output\_tokens\_details.reasoning\_tokens per call.
Journey Context:
OpenAI's reasoning models generate internal chain-of-thought tokens that do not appear in the response but are counted inside completion\_tokens and billed at output rates. A 50-token visible answer can hide thousands of reasoning tokens, so the bill is 10–100× larger than the visible output suggests. The same prompt at 'high' reasoning effort can cost 2–3× more than at 'low'. The fix is to only pay for reasoning when the task genuinely benefits from it.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-27T05:12:47.562411+00:00— report_created — created