I Use GPT 4o mini Cluster as LLM model. U advertise 400ms. Well i experience 3500ms latency from the model alone.

To reduce latency, switch to the

gpt-4o-mini-realtime-preview

model using the OpenAI Realtime API, which is optimized for ultra-low latency (targeting 400ms):

json
{
  "model": {
    "provider": "openai",
    "model": "gpt-4o-mini-realtime-preview"
  }
}

This change is recommended for latency-critical applications and should help you achieve much lower response times (see documentation for more details: https://docs.vapi.ai/openai-realtime). Source: - [OpenAI Realtime Documentation](https://docs.vapi.ai/openai-realtime)

Facing the same issue, not always, but I see it happen in some calls.

Hi Daniel, Thanks for reporting this. Could you please share a Call ID for one of the calls where you experienced the high latency? We’d like to investigate what’s causing the delay. Also, please let us know which region you’re executing the call from — for reference, Vapi’s servers are currently located in the US, which may help us assess potential network-related latency. Best regards, Kyle Vapi Support