is it me or has vapi latency massively improve. I remember it being so bad that I couldnt use for any client work despite really liking it but now even for bigger prompts and model that latency is insanely fast. Makes it so much more human like