Presumably because OpenAI charges per-request fees. These services aren't training an in-house model, but are leveraging an LLM via an API and they have to pay

Presumably because OpenAI charges per-request fees. These services aren't training an in-house model, but are leveraging an LLM via an API and they have to pay those fees. (I'm not saying this to imply that building a service like this is simple just because an API is used - there's still a lot of plumbing work and designing a way to manage document embeddings, etc).