OpenAI sending a direct 500 Internal Server Error (for some clients)

API Bugs
1

Hello Dear Team

I have some clients (<10) but in the past hour and all after a release we made earlier this week, came back that they are getting a 500 error from openai.

this is the flow

  1. The AI Agent initializes with the configured realtime model (gpt-realtime-2).

  2. It successfully authenticates and mints a client secret session via POST https://api.openai.com/v1/realtime/client_secrets (returns 200 OK).

  3. It then attempts to initiate the realtime WebRTC call by sending an SDP offer to POST ``https://api.openai.com/v1/realtime/calls.

  4. OpenAI responds with a 500 Internal Server Error:
    2026-07-31 10:11:18.655| Info|{‘call’: 17, ‘dir’: ‘ai’} OpenAI response: https://api.openai.com/v1/realtime/calls (500) http_version=HTTP/2 openai_call_id=/v1/realtime/calls/rtc_u2_E7iFCkVxgC7oMS7P7HEsU request_id=None cf_ray=a23d31088d4c7411-IAD

    2026-07-31 10:11:18.656|Error|{‘call’: 17, ‘dir’: ‘ai’} realtime/calls error body: retry_after=None body=Internal Server Error

    2026-07-31 10:11:18.657|Error|{‘call’: 17} [FATAL] AI Session entered fatal error state: Failed to start AI session: Server error ‘500 Internal Server Error’ for url ‘https://api.openai.com/v1/realtime/calls’

  5. Because OpenAI returns a 500 Internal Server Error without any parsable JSON explanation (just the raw Internal Server Error text body), the 3CX AI bridge correctly marks the session as a fatal error and triggers a fallback transfer.

    Am I sending this correctly?

    "audio": {
    "input": {
        "turn_detection": {
            "type": "server_vad",
            "threshold": 0.8,
            "prefix_padding_ms": 500,
            "silence_duration_ms": 700,
            "create_response": false
        },
        "transcription": {
            "model": "gpt-realtime-whisper"
        }
    },
    "output": {
        "voice": "marin"
    }
}

    Im concerned about this line because this line is the difference between the previous version and the new update we made. “model”: “gpt-realtime-whisper”. (in the previous version it was whisper-1) But this works perfectly in our environment and we made around 5k calls per day in evals for 1 week before our release.

    Can you please help me understand if this could be temporary or not?

    Thanks so much and at your disposal for anything you need.

    God bless

2

Client informed me that without doing nothing it just resumed working.

We can mark this as solved.

3

Hi Dear Team

We got another report for a customer. Sometimes openai answers with 500.

  1. Successful PBX Connection: The call itself connected successfully to the PBX (at 08:32:44.420).

2026-08-06 08:32:44.420| Info|{‘call’: 6} Connected

  1. OpenAI Failure: We try to establish a voice session with OpenAI by making a request to realtime API (https://api.openai.com/v1/realtime/calls). However, OpenAI responds with a 500 Internal Server Error.

2026-08-06 08:32:44.664|Error|{‘call’: 6, ‘dir’: ‘ai’} realtime/calls error body: retry_after=None body=Internal Server Error

2026-08-06 08:32:44.664|Error|{‘call’: 6} [FATAL] AI Session entered fatal error state: Failed to start AI session: Server error ‘500 Internal Server Error’ for url ‘https://api.openai.com/v1/realtime/calls’

For more information check: 500 Internal Server Error - HTTP | MDN

In the log the client gave me there are tens of hundreds of calls that worked. Suddenly this happens. There are no changes refreshed between the working calls and the failed calls. (no changes of model, no updates, no configuration changes)

When the API functions correctly, OpenAI responds with an HTTP 201 (Created) status code instead of a 500. I found over 100 successful connection attempts throughout the log file.

This particular failure at 08:32 just appears to be an unlucky moment where OpenAI’s realtime voice API briefly went down or experienced an internal glitch. The previous call was just a few seconds earlier and that was successful.

Successful call: 2026-08-06 08:31:52.909 (Line 21426)

Failed call: 2026-08-06 08:32:44.664|Error|{‘call’: 6} [FATAL] AI Session entered fatal error state: Failed to start AI session: Server error ‘500 Internal Server Error’ for url ‘https://api.openai.com/v1/realtime/calls’ For more information check: 500 Internal Server Error - HTTP | MDN

Can some one shed some light on this please - any help will be greatly appreciated.

God bless