Gpt-4-turbo slightly worse than gpt-4-turbo-preview when generating text based on images?

I have a prompt for generating documentation content starting from some initial text, images and instructions.

In my tests, the new ‘gpt-4-turbo’ performed sligtly worse overall when compared to ‘gpt-4-turbo-preview’. At least that was my feeling. There were also instances where it performed better, but overall I was a bit dissapointed with it (maybe because I expected it to be even better than the preview model).

For example, it created a list with only one bullet where the preview model rightly generated a paragraph. Or it didn’t create the expected XML markup (based on an example), where the preview model did create it.

1 Like

In my experience, it does not follow my directions (on formatting and citation for text generation) nearly as well as -preview model. I reverted back immediately.