I have set up a test. Specifically, the task is to rewrite existing texts. I compared the results of the API with the results directly from the ChatGPT interface. The GPT 4o model and the prompt are identical for both.
The result: The texts rewritten by ChatGPT are much better - the API shortens the texts in the rewritten version excessively, which means that a lot of relevant information is lost.
My question: Why is this the case?