Noticable difference in instructions following of systemprompt in assisstant api vs chatgt 'custom GPT'

I noticed that the instruction following from the custom created gpt differs
from the assisstant api.
Gave the same systemprompt and query to both.
Chatgpt rarly follows the instrucions in a clear way where the assisstant api in most cases follows its very acurate.
I turned off the custom user instructions in chatgpt so it should only follow the given instructions in each gpt prompt.
Any explanations for that ?

Hi and welcome to the Developer Forum!

It is my understanding that the assistant system makes use of the system prompt and GPTs do not, if you put the instructions into the custom prompt you may experience a more similar result, worth a test.