GPT-5 is pretty fast at web searching (like instant). It seems to pass those requests to the nano model which means it's always going to miss the fucking point if you ask it to search a topic on the web for you. Brilliant. But it was kind enough to plug a bunch of shitty normie ezines at the end. (Which happen to be what it sourced. It didn't even try to search any tech forums).
This of course after another query where I asked it to make a comic about me cancelling my ChatGPT plus subscription over the removal of the model selector- thus there should have been vector memories for it to draw on. But I guess nano doesn't do that.