I am building an AI voice agent using VAPI and Make.com. An AI voice agent does booking, rescheduling, and canceling of appointments. It also provides an FAQ and business info. Now the issue is that that VAPI agent does not strictly follow instructions. So can you tell me which configuration works? For voice using the VAPI provider. If you need more info let me know.

For an affordable and instruction-following AI voice agent, use the

gpt-4o-mini-realtime-preview-2024-12-17

model with a clear system message and a supported voice. Here’s a sample VAPI configuration:

json
{
  "model": {
    "provider": "openai",
    "model": "gpt-4o-mini-realtime-preview-2024-12-17",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant. Always follow instructions exactly. Confirm actions before proceeding."
      }
    ],
    "temperature": 0.6,
    "maxTokens": 250
  },
  "voice": {
    "provider": "openai",
    "voiceId": "alloy"
  }
}

This setup is cost-effective and optimized for strict instruction adherence. Test and iterate your prompts for best results. Source: [Vapi OpenAI Realtime Documentation](https://docs.vapi.ai/openai-realtime)

what is the pricing of gpt-4o-mini-realtime-preview-2024-12-17 this model in vapi?

Actually, we recommend using gpt-4.1-mini or gpt-4o-mini for a cost effective, instruction following assistant using deepgram as transcriber and deepgram/elevenlabs/vapi for voice. For a more cost effective voice option, you can use azure.