Deepgram nova2/3, groq llama 3.3 70b Versatile, llama 4 scout or openai 4.1-nano as a Custom LLM, 11labs flash. I'm in Europe so roundtrips are quite long, but its usually ~800ms - 1200ms