Gpt4 was doing it's job though, it's the work around I found for the problem. Are you using 3.5 just for latency purposes?