Hello,
I was working on extracting some codes and numerical data from images with long text and graphs. Since many exist, I need to get the output in a structured format and decided to go for “gpt-4o-2024-08-06”. The problem is that the model is completely fabricating the numbers in the given images. Whether it’s low or high detail selected, it still completely fabricates the output. Is there any way to get over this? I wished to use older versions, but the speed and unstructured output constraints blocked me.
Your advice is highly appreciated! Thank you.