How to reduce the latency of the streaming avatar?

I am using the OpenAI 4o API, but the avatar's response is taking about 5 seconds to process. I'm considering switching to Gemini Flash 1.5, which provided a response in around 2 seconds. Is this the best solution?

Or is there another way to improve latency while continuing to use OpenAI?

I saw that you released the new Realtime API (Heygen), but it's still in the testing phase, right? I'm concerned about using it and then it being discontinued.