I don't recall the original version we used unfortunately :( in our case, the bump was actually from gpt-4-vision to gpt-4o (the use case required image interp

I don't recall the original version we used unfortunately :( in our case, the bump was actually from gpt-4-vision to gpt-4o (the use case required image interpretation) It got measurably better at both image cases and text-only cases