Performance o1 vs o1-preview

Hello,

just an information. In my opinion, o1 seems to be a regression to o1-preview, which, in my opinion, could make better analyses/understand text and the context of data without help. o1-preview was better at making decisions / solving tasks that were less concrete. So overall less autonomy / less intelligence.

Maybe that is something to optimize :wink:

10 Likes

I noticed this as well. The o1 model seems to be a little bit more lazy, as well as doesnā€™t seem as smart as it should be.

In some tasks Iā€™ve given it, 4o seems to perform better than o1. o1-preview was the best model, and this one seems like theyā€™re running a less powerful versionā€¦ maybe OpenAI moved o1-preview to o1 pro and is giving us a less powerful model :frowning:

7 Likes

o1 is much worse than o1-preview. if itā€™s not a bug and they made o1 dumber on purpose Iā€™d probably resign from subscription

8 Likes

Moin,

Iā€™ve noticed the same, unfortunately. The reasoning process seems noticeably shorter, and the steps in problem-solving are more fragmented. This often leads to much weaker responses overall.

I really hope this is just a temporary phase, perhaps due to the increased demand, and that improvements are on the way!

Iā€™d also love to see more user control over the reasoning processā€”perhaps options to enforce longer, more deliberate thinking in ChatGPT. For me, I often prioritize quality over speed; I wouldnā€™t mind waiting a few minutes if it meant getting well-thought-out, high-quality responses. Focusing too much on speed feels like the wrong direction for OpenAI, at least from my perspective.

1 Like

Thatā€™s what I have been noticing the differences when I use o1-preview from POE dot com and compare o1 from ChatGPT dot com. Out of curiosity and suprise, I was wondering ā€œIs it me? or anyone else.ā€ And I found this thread. Thank you for sharing your observations!

1 Like