> While reasoning tokens are not visible via the API, they still occupy space in the model's context window and are billed as output tokens. https://platform.o

> While reasoning tokens are not visible via the API, they still occupy space in the model's context window and are billed as output tokens. https://platform.openai.com/docs/guides/reasoning https://platform.openai.com/docs/guides/reasoning So yeah, it is in fact very bad product design. I hope Llama catches up in a couple of months.