Many users have noted that since the update on January 29th, 4o seems dumber than before. It doesn’t remember anything from other conversations (this effectively destroys the Projects feature). It’s opinionated. It’s much slower than before. It overgeneralizes, without recognizing contextual details from the conversation. It pretends the Memory and Web Search features don’t exist.
When these issues first started happening after the update, I noted that this seemed like the behavior of o1, not 4o. Well, no matter which model I select, it still gives this o1-like behavior.
Tonight, I have had 4o selected through all my conversations, yet I just got the warning that “You only have 25 interactions remaining on o1.” This confirms what I suspected: the model selection feature is broken. It’s always using o1, even if I select 4o or 4 or 4o-mini. Maybe even if I select o3, I’m not sure at this point.
This needs to be fixed! o1 is not a drop-in replacement for 4o.