gpt 4.1 mini, deepgram, elevenlabs v2 flash. I'm using this stack and the latency is great, plus the assistant quality is great too.