Is there any know issue with gpt-4o-mini, using a knowledge database.
I have some recipe in the knowledge database, when I use gpt-4o-mini, most of the time it return, I don’t have a recipe for that.
Using O3-mini work just fine.
gpt-4o-mini use to work with no problem, just in the last couple of days, it doesn’t work well
Yes there are API issues since some hour at least with GPT-4o-mini. We have a project running with multiple requests per minute. Many just won’t get a response or get fulfilled. So there seems to be instabilities. In OpenAI Status there is also something highlighted: https://status.openai.com/ but not specifically to this issue.
@tom135 you using batch processing or normal API calls for that? I’m seeing issues with batch and wanted to see if your was related
I’m using Normal API calls
I’m using batch processing with apiconcurrency: 20. need to manually check every hour here and there. today had to start again 3 times already. sometimes also had to delete parts as it just left out certain batches and continued elsewhere. not sure what is going on. how is it for you?
Looks like batch processing is having issues based on this thread - Batch api stuck on `Validating` status? - #26 by kevin58
gpt-4o via batch, none of the jobs are leaving “validating” stage. I have opened a thread with support and will update both threads when I have any hard info.
we got a response on the other thread - looks like internal issue was fixed related to batch jobs - Batch api stuck on `Validating` status? - #27 by andyw1
What is also weird is that 0 incident are reported on the APIs, batch or no batch. I’m not sure we can trust the status page to give real status
For me the issue is still the same gpt-4o-mini doesn’t work well right now
I’m running constantly batches 24/7 - the issues remain.
I also looked up status pages. Nothing to see there. I could see that there is a dependency on output token. Batches with low volume of output token had no issues, while higher output token volumes got hiccups and needed manual restart of batches.
Seems a very specific problem with the API use of 4o-mini imho.