How do you know “the exact setup” is truly “exact”? I bet custom GPTs have some extra system messages going on, and maybe some extra steps in answering process, which affect the results. Unless you know all of what is going under the hood on both sides, it’s kind of shaky to assume the setups are exactly the same.
@sergeliatko I find this is by far the simplest answer between assistant’s api vs completions. If you want to update your response to add this nuance - assistants api - comes with a vector store but you as a developer don’t have access to it. completions api - can be hooked up to any external vector db. I also encourage this to be a medium/substack article for broader sharing.
1 Like
@kavitatipnis thanks for the suggestion, very useful. I used chat to update the response (I confirm it is what I think)