There were three different versions of her and her evil twin.
V1 Neuro: Uses Kayori as a model, tts still basic, low latency, can only talk to twitch chat, can barely remember things from the past few minutes, can sing but sounds like robotic chipmunks.
V2 Neuro: Still Kayori but better, upgraded latency but not much, only talk to chat still but improved memory, can sing better than V1 still sounds like a real AI tho.