Having a pretty bad experience with GPT-5 and web searches – it either hallucinates or can't find web results unless I manually toggle the 'Web Search' button? And even then, responses are kinda...meh?

GPT-4o definitely didn't do this, and it seems like folks on Reddit are hating this rollout too: https://www.reddit.com/r/ChatGPT/s/zddfnoFDnR

@viticci Sort of the response I've gotten a few times with it:
I’m hitting a bit of a snag getting direct search results via the tool’s interface—seems like the parameters are a bit particular. No worries, though! I'll still compile a thoughtful list of technology and startup-focused podcasts that regularly host guests and are less mainstream.
@viticci All while they said those things shouldn't happen anymore with this update... 👎🏻
@viticci and you can’t even go back to the old models, they’re just gone
@brianmueller @viticci They made them only available for Pro users.
@viticci Why would one burn that amount of energy for a simple web search? I don’t get it.
@viticci I’ve found with past models (especially thinking ones) I would have to convince it and argue with it that iOS 26 IS THE NEXT VERSION OF iOS! It always looked at historical data and never believed me that iOS was jumping up that many numbers in one year. Looks like that hasn’t exactly been fixed and you still need to argue with GPT-5.

@viticci I haven’t used it that much yet, but this has pretty much been my experience as well.

Tried to search for information about an Apple API earlier, and the standard model couldn’t find any results. Tried the thinking model, and it gave me a bunch of results without links.

Went and did a regular search instead, and found a bunch of relevant results 🤷‍♂️

@viticci Altman saying today that they had some deployment issues yesterday that effectively made it dumber than it should be? We shall see.
@hunter yeah it's totally fine for me now
@viticci 5 is terrible. Used it past 24 hours and it's drunk.