Are you based in US? I found Open Ai latency much worse in Europe. So far groq/llama works best (latency wise) - buts that’s still just testing on web - vonage/twillio adds another 200-250 on top