Is GPT4-o dumber in Assistans API than in normal chat?

[tobson]

Hello,

I am asking myself if the new gtp4-o model ist working differently via assistants API than via the casual chat in the browser?

I would like to use the vision capabilities to transcribe handwritten notes. If I put a photo of my handwritten notes into ChatGPT gpt-4o it perfectly transcribes my ugly written notes. But if I am creating an assistant in the playground (of course using the gpt-4o model too) it gets a lot of things wrong and starts to hallucinate.

I tried it a bunch of times and the result is always the same: Via the normal chat it works perfectly, in assistants API it works badly.

How can that be? Isn’t it the same model, so the results should be the same?

Thanks!

[mariayeruh]

Got the same behavior in the same tasks.

GPT-4 omni text recognition via API works worse than on chatgpt_com

Hello all! I’m investigation poor text recognition via API with GPT-4 Omni. The original OpenAI chat on chatgpt_com is working like a charm, text is 100% equal to PNG, no fictional words or sentences. If I us API call to GPT-4o for a one page text I always get only the first paragraph almost correct, the others are fictional. I tried custom prompts to stop using Tesseract and use internal vision capabilities. But no luck. What should I do? [proof]

[jonathan83]

Hi, I am facing to the same problem.
Will you find the reason or a workaround?
Thx

[the.sliwinski]

Same here, the performance of vision in assistant is significantly worse compared to classic gpt4o.