My understanding is that GPT-4o is acting as a text LLM right now. It sounds like OpenAI will be releasing the voice mode soon and hopefully VAPI will take advantage of it eliminating a few stages and reducing latency.