over the past 4-5 days, i started noticing that chatgpt had become noticeably dumber, but i did not have time to look into it. today i became 100% sure of it.
i need chatgpt for work. it is an excellent assistant for solving my work tasks. i add information to the sources, and with that information my work gets done twice as fast and with better accuracy.
information, accuracy, and correctness are important to me. today i wanted to update my sources and added more of my own study materials, but chatgpt was doing it WAY too fast. i have been using chatgpt for a long time, i can feel how it thinks, and this time my gut feeling did not let me down. it really was answering way too quickly without even thinking. i tested it on several work-related tasks, and it really started missing A LOT of important details that it never used to miss before.
i think you know that feeling when the usual rhythm gets disrupted and something just feels wrong. in my case, it was genuinely noticeable.
in the first screenshot, chatgpt says that it is currently running on the gpt5.5 mini model, which, as far as i know, DOES NOT EVEN EXIST.
after that, i started pestering it about why it was running on this model and how i could fix the problem. the ui clearly had gpt5.6 sol selected.
then i looked online and found information saying that the number of active sessions affects how chatgpt performs, so i cleared out the extra ones.
and it worked! in the second screenshot, you can see that chatgpt identifies itself as gpt5.6 sol, exactly as it should. so the argument that “chatgpt does not know its own model version” does not really hold up here, because i tested the exact same question with the chatgpt o3 model, and it correctly identified its model version.
for now, it seems that what helped was me actively looking for ways and workarounds to compare the model version shown in the ui with what i was actually getting. it almost feels like chatgpt realized that i was starting to suspect something and quietly switched everything back to normal.
i am asking everyone who has run into this problem to test it and report your results. my method might not be 100% reliable, it is just a diagnostic approach that, for some reason, produced results.
there is no official information from openai about this anywhere.
grace be with you!
after (prompt full same with last screen):
