Get used to it, these models aren’t going to replace fast programming. They have to think for a long time to gaslight you into thinking it produces good code. Learn to program.
I tried GPT-5 on vscode, and it is unbearably slow. So slow that it is not usable.
And that last commenter who said “get used to it, that’s how it is” - that’s wrong. Cloud code is pretty good at what it does, and is fast in comparison. To put some numbers on it, I would say Claude Code takes 5-8 seconds per prompt, unless you ask for research and hard thinking. For the same basic prompt, vscode with gpt-5 takes 3-4 minutes. I am using a new MacBook Pro (M4 Pro).
Am I the only one that understands that a gazillion people are playing with gpt-5 right now as it rolls out, so it’s going to be slow whilst OpenAI assesses demand and provides capacity for it.
This is free. For us to try. It’s potentially the most capable AI ever produced by mankind and people are whining it’s a bit slow?!
Jeez, we’ve become an entitled bunch. We’re witnessing the most incredible shift in technology, and getting to use it in real-time as it iterates towards forever-changing how humans interact with data, each other and the world.
But yeah, it’s currently a bit slow…how disappointing.
I don’t know about others, but I mostly program (fast) manually, but Cursor has become a great assistant, and with that said, the original bug report of GPT-5 still holds true. Do you use Cursor?
Gpt-5 is slower than the other models Cursor was using for Auto regularly (GPT 4.1 and Sonnet 4). It is objectively behaving differently than those models, and even with brain dead easy questions that the free google search AI could answer. It wouldn’t be a big problem if it wasn’t forced when using Auto where I can’t choose which model is used.
It is disappointing when it is forced on us who are using Auto and we are used to Auto not taking 3-4 mins per request. People have integrated Cursor into their daily work, and it is a big change when suddenly requests take 3-4 mins instead of 10-20 seconds.. I am not choosing GPT-5, Cursor is choosing it for me via Auto.
Update: it seems a bit faster, maybe it was just the first few requests that it was very slow.
@MidnightOak Auto may not be related to GPT-5, we had infrequent reports of Auto being slow on some requests even before GPT-5 release. let me know the Request ID if you catch it again.