Is anyone else's GPT-4o and o1 suddenly acting... dumb? 🤔

[1601344864]

Hey Plus users, am I going crazy or did OpenAI secretly swap our models?

Been using Plus for months, but recently noticed:

GPT-4o feels like a mini version
Can’t use Code Interpreter anymore (???)
GPT-o1 lost all reasoning ability
Still paying $20/month
Quick test: Ask it to solve any complex problem. Notice how it doesn’t show ANY reasoning steps anymore? Try Code Interpreter - suddenly “not available”?

The weird part? No announcements. No emails. Nothing.

So either:

My account is bugged
OpenAI quietly downgraded our models
I’m just imagining things
Anyone else notice this? Would love to hear your experiences.

PS: Has anyone successfully gotten a refund after noticing this change? Asking for a friend

[xbcdr]

Same here, but it seems like no one cares

ChatGPT model random downgrades

When using the GPT-4o model, I’ve been encountering frequent refusals to answer: “I’m sorry, I can’t assist with that …” Once the model refuses, its responses drop significantly in quality, and this can last anywhere from several hours to a full day. One time it even refused to answer a question about the structure of a car engine wtf. So, I switched to the o1-preview model, only to find that it gave an immediate answer without any thought process. When I asked if it was the “o1” model, it re…

(5 views in 5 hours)

[j.peero]

i was using it this morning normal, and it has since turned straight stupid. i will ask it something and it will bring up something completely irrelevant.

when i tried using the conversation function on my phone, it did not work at all.

[chengshi797]

me to ,the responses from the model are too brief and useless now

[wynistccz]

I’ve also encountered this issue. Previously, when I used it on the Windows browser, this problem frequently occurred, but it was normal on the Android app. However, since yesterday, this issue has also emerged on the Android app. All regular models (gpt-4, gpt-4o) have completely degraded to the quality level of gpt-4o-mini. o1 and o1-mini either fail to reason properly, or if they do reason, the quality of the generated output is terrible.

[dignity_for_all]

The models used in ChatGPT are not entirely identical to those provided via the API.

For GPT-4o, the model used in ChatGPT corresponds to gpt-4o-latest in the API, but even so, the responses might sometimes come across differently from what is seen in ChatGPT.

Regarding the o1 model, it does not fully disclose its thought process (Chain of Thought, CoT) but only provides a key summary. There may be some design considerations behind this approach, possibly to limit exposure of internal mechanisms. Recently, it appears that there are also cases where the CoT summary is not displayed at all.

In any case, since ChatGPT operates on a subscription model, it needs to ensure that limited computing resources can be used by as many people as possible. For this reason, frequent adjustments to the model’s behavior are not something that’s intentionally hidden.

If the experience falls significantly short of expectations, you may have the option to request a refund by reaching out to support. However, ChatGPT’s performance fluctuates frequently (sometimes improving and sometimes worsening), so it could be helpful to occasionally check whether it meets your expectations.