On OpenAI demo, the AI seem to have a crazy low latency, however I don't see the same happening while Using Vapi and Gpt 4o, does anyone know why is that?