Not necessarily. Input tokens is far cheaper than output tokens. The alternative is to have the LLM derive the documentation from code each session, which is expensive due to the amount of output tokens it produces.