What i liked is that vapi TTS models are getting good, connectivity is pretty good, since it's not something that heavy. Latency on the other hand is where vapi should start looking at, before the call, it says the latency is 1.1s for example but post call the time per turn takes 2s+