500 Internal Server Error | Deepgram | Page 1

boreal juniper Oct 13, 2025, 6:43 PM

#

Was there an update or change over the weekend? Friday all of my voice agents were working without issue. Today we came in with customers complaining and I'm getting random 500 Internal Server Errors from Deepgram, like so-
{'type': 'Warning', 'description': 'Error making request to think provider. Code: 500 Internal Server Error. Body: {"error":{"code":500,"message":"The server had an error processing your request. Sorry about that! You can retry your request, or contact [email protected] if you keep seeing this error.","param":null,"type":"cr_error"}}', 'code': 'THINK_REQUEST_FAILED'}

The odd thing is, it's happening at random times without even sending and data/STT from the client.
I've been troubleshooting my function calls thinking they are the issue but as I dig deeper, it's happening outside of function calls.
Here's my latest settings
{"type": "Settings", "audio": {"input": {"encoding": "mulaw", "sample_rate": 8000}, "output": {"encoding": "mulaw", "sample_rate": 8000, "container": "none"}}, "agent": {"listen": {"provider": {"type": "deepgram", "model": "nova-3"}}, "think": {"provider": {"type": "open_ai", "model": "gpt-4.1-mini", "temperature": 0.7}
I changed from 4.0-mini to see if the model was the problem and I have NOT tried another provider.
Here's one request Id for reference but pretty much every one in my account is doing this today. 45452d54-c3ed-44a5-a560-b814aef06556

snow quiverBOT Oct 13, 2025, 6:43 PM

#

Thanks for asking your question. Please be sure to reply with as much detail as possible so the community can assist you efficiently.
-# If you haven't done so, ensure your Discord and Github profiles are linked to Deepgram so you can earn points to redeem on cool stuff just by being active!

#

It looks like we're missing some important information to help debug your issue. Would you mind providing us with the following details in a reply?

The programming language you are working in (e.g. JavaScript, Python).
The deepgram product you are using (e.g Speech to Text, Agent API)

boreal juniper Oct 13, 2025, 6:50 PM

#

Python, Agent API

twin umbraBOT Oct 13, 2025, 6:53 PM

#

Give the error states:

The server had an error processing your request. Sorry about that! You can retry your request, or contact [email protected] if you keep seeing this error
It would appear to be an upstream ongoing issue with OpenAI's services. We've been seeing intermittent issues for a day or so. Please switch provider if it isn't going to cause too much impact on your users

boreal juniper Oct 13, 2025, 7:17 PM

#

I apologize, was reluctant to try another provider because of the speed but I suppose I should have. Any recommendation to save me some testing and cost? Anthropic is my go-to for other AI applications but the speed of responses on the larger models turned me away during beta testing for voice applications. I have not tried any of them since Agent's came out of beta.
4.o-mini also gave the most consistent function calls if I recall but at the time my only other choice was Haiku-3.5.

brave saffron Oct 13, 2025, 9:41 PM

#

Note that we're seeing the same with gpt-4o-mini. Not having this issue with gpt-4.1-mini, although I don't the like the different behavior we get out of this model with our Voice Agent.

boreal juniper Oct 14, 2025, 11:37 AM

#

I still get the same results with 4.1-mini. Other providers just have difficulty running functions consistently and that's mostly what our agents are doing. Is there any engagement with OpenAI to get this resolved? I did find a lot of posts in their forum from last week about the Assistants API returning the same 500 error but it seems everyone is saying that was resolved.

brave saffron Oct 14, 2025, 10:30 PM

#

These problems are continuing for me today, and I'm seeing the same errors now with both gpt-4o-mini and with gpt-4.1-mini. Also seeing the dreaded SLOW_THINK_REQUEST warnings as well. The Deepgram status page and the OpenAI status pages are all green, but it sounds like this is a known issue with Deepgram. Is anything being done about this? If my Voice Agent were in Production we'd now be on day two of what would be a pretty serious outage.

boreal juniper Oct 15, 2025, 11:37 AM

#

I received an email from Deepgram support last night confirming the issue and said they have engaged OpenAI to try and get it fixed. Unfortunately we have a few in production that are now down which is obviously concerning going forward. Changing providers is not a simple fix, they just don't work the same.

brave saffron Oct 15, 2025, 12:35 PM

#

I've also reached out to our reps to try and get some help here. I totally agree with the provider differences. We're finding that anything complex prompt-wise requires model-specific prompts almost always. Significant differences even between gpt-4o-mini and gpt-4.1-mini, for example. I can't imagine that any useful Voice Agents would have simplistic prompts. We're exploring segmenting our prompts up into smaller sections using UpdatePrompt as we move forward in a flow.

boreal juniper Oct 16, 2025, 4:07 PM

#

@brave saffron we applied a workaround until we can change code to use a new vendor all together. in your agent.think settings add "endpoint" and use OpenAI's completion api directly-
https://api.openai.com/v1/chat/completions
it should look like this-
"endpoint": {
"url": "https://api.openai.com/v1/chat/completions",
"headers": {
"authorization": "Bearer OPENAI_API_KEY"
}
}

we've been testing our agents today and all models including 4o-mini are working fine direct so it's a deepgram problem, not OpenAI.

brave saffron Oct 16, 2025, 6:02 PM

#

Great idea. I'll give that a go. Thank you so much for the tip! Living on the bleeding edge is always exciting!

#500 Internal Server Error