Yep as soon as query per second matters, and cost per query... Definitely going through that on some work now, where we use GPT4 on subset and rest on GPT3 or t

Yep as soon as query per second matters, and cost per query... Definitely going through that on some work now, where we use GPT4 on subset and rest on GPT3 or tuned. Only slow user-facing and other low-volume / high-latency goes to GPT4..