ratty-blush•2w ago

GPT-5 mini randomly ending calls

GPT-5 mini uses the Pre-defined "end call" tool in a weird way, ending the call on the first message or mid-conversation, when it's definitely not appropriate to. Which makes the best low-latency model so far unusable. GPT-4.1-mini in the same prompt never had such a problem in over 10k calls. I suppose it has something to do with how this Pre-defined function is manifested in the backend, and how GPT-5 mini handles it :vapilogo:

8 Replies

Vapi•2w ago

Suggestion:
To prevent GPT-5 mini from ending calls inappropriately, review your assistant's system prompt and tool configuration. Make sure the endCall tool is only included when necessary, and update your prompt to avoid mentioning or implying call termination except in explicit scenarios. For example:

{
  "model": {
    "provider": "openai",
    "model": "gpt-5-mini",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant. Only use the endCall tool if the user clearly requests to end the call or says goodbye."
      }
    ],
    "tools": [
      { "type": "endCall" }
    ]
  }
}

{
  "model": {
    "provider": "openai",
    "model": "gpt-5-mini",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant. Only use the endCall tool if the user clearly requests to end the call or says goodbye."
      }
    ],
    "tools": [
      { "type": "endCall" }
    ]
  }
}

Tip: Remove the endCall tool from the tools array if you want to fully disable this behavior.
See more in the Vapi documentation. Sources: - Default Tools – End Call - Prompting Guide – Error Handling & Tool Use

⤵️ Mention @Vapi Support Bot for follow-ups 🐥

foreign-sapphire•2w ago

I'm having similar issue with GPT-5 and GPT-5-mini, it ends the call as soon as a user speaks. I've removed my global prompt entirely It would be great to get some help with calls ending unexpectedly for GPT-5* models. Switching to 4.1 or another model, the issue is not present. I have cleared my global prompt to make sure that wasn't causing it to end early...

Shubham Bajaj•2w ago

If your calls are randomly ending while using GPT-5 mini, it's helpful to check for specific call ended reason codes in your call logs. Common reasons could be assistant-related issues or errors in the AI processing pipeline. You can refer to the call ended reasons documentation for more detailed codes and their meanings. Additionally, if you're seeing frequent logouts or error messages, it might be an issue related to your authentication or account settings. Please ensure that your login credentials are up-to-date and check the account security settings for any irregularities. For detailed error codes or if the issue persists with specific behaviors, providing a Call ID could help in further diagnosis.

Call ended reasons | Vapi

All possible call ended reason codes and what they mean.

stormy-gold•2w ago

I have the same issue. I can use gpt-5-nano and other models in Vapi without any problem, but gpt-5-mini and gpt-5 don’t work. The calls end abruptly and I get this error: "pipeline-error-openai-400-bad-request-validation-failed".

Shubham Bajaj•2w ago

To help track down this issue, could you share: - The call ID -
This would really help us figure out what went wrong!

stormy-gold•2w ago

Call 1: 8d08cdc8-e85f-4a6d-ad19-2439b0aa712b I see now in my call log details it showed this message: Model request failed (attempt #1, reason: Your organization must be verified to stream this model. Please go to: https://platform.openai.com/settings/organization/general and click on Verify Organization. If you just verified, it can take up to 15 minutes for access to propagate.) (provider: openai, model: gpt-5, region: unknown, credential: true) I've gone ahead and asked my client to do so. I will update you after it's done and let you know if this solution worked I can confirm this solved the problem for me

ratty-blushOP•2w ago

Oh, so you used it via OpenAI's API. And do you have a problem with GPT-5-mini randomly ending calls? Or it works smoothly now?

Shubham Bajaj•5d ago

If you are using Vapi's default API key, you wont encounter the organization verification issue since Vapi is already verified to use these models. AI ID if you are still encountering issues with your calls ending prematurely please share a call id where it occurs and we will review the logs

GPT-5 mini randomly ending calls

Did you find this page helpful?