Inconsistent Performance of 4o-mini Model in Claygent Tasks

·May 21, 2025 02:25 PM·

Share

Hi everyone, has anyone had any recent bad experiences with 4o-mini inside of Clay? especially for Claygent tasks? I usually use 4o-mini for simple and short tasks since it is cheaper and also works but recently I found that it is not consistent enough. For example: If I try to check if a website is not working with the help of Claygent and using 4o-mini as a model with my own API key, it has a 1 out of 4 error rate. While 4.1 mini is almost 100% accurate (with the same prompt), I was wondering if something changed inside OpenAI's model limitations and I am not aware of it. I wanted to ask the community here first if any of you had similar experiences with the models especially after the new ones came out few weeks ago. Thanks

2 comments

· Sorted by Oldest

kushagra t.
·
·
4.1 mini def has been performing better for me as well but don't see a lot of regression with 4o mini either recommend using json output in case you aren't already and checking if that reduces the bad results
1
Justin R.
·
·
4.1 mini is better for well structured prompts. just keep in mind even the best prompts are 90% accurate, so you need to build validation and re-try logic.
1

kushagra t.
·
·
4.1 mini def has been performing better for me as well but don't see a lot of regression with 4o mini either recommend using json output in case you aren't already and checking if that reduces the bad results
1
Justin R.
·
·
4.1 mini is better for well structured prompts. just keep in mind even the best prompts are 90% accurate, so you need to build validation and re-try logic.
1