Models can't code anymore | OpenAI | Page 1

worn robinBOT Apr 20, 2025, 3:40 PM

#

Reported by @urban lodge

Bug Report: Models can't code anymore

`Steps to Reproduce`

I checked all models, none of them can stay on task coding anymore, they all cap out at around 200 lines of code

`Expected Result`

I expected it to work like before, the 4o model could do over 1000 lines of code a week ago.

`Actual Result`

Now all models cap out around 200 lines of code.

`Environment`

All platforms

#

Additional Information

Please provide relevant details to help resolve the issue, such as:

ChatGPT Shared Link (if applicable).
Screenshots or videos demonstrating the problem.

-# ➜ Need to contact support? Visit the OpenAI Help Center.

urban lodge Apr 20, 2025, 3:41 PM

#

They can't handle context anymore as well, you tell it one thing, it does it, you tell it another thing, it forgets about the first thing

silk marsh Apr 21, 2025, 1:56 AM

#

Discussing with others who agree. The new models are garbage.

I am using o1-pro still but I'm pretty annoyed that it was put into Legacy considering the new models don't come remotely close.

If this is the direction OpenAI is going and all these days without follow-up, it's obviously time to jump ship.

urban lodge Apr 26, 2025, 2:22 AM

#

Just noticed today that 4o model instead of replying with code, will reply that its about to tell me the code in 20 seconds then ends message with no code, I have to then ask for the code again to get the code. Double the messages for each thing, I actually hit 4o limit today for the first time.

iron lance Apr 26, 2025, 4:09 AM

#

It's become astoundingly dumb. I've been working on some frontend redesign work (Razor/MVC/C#) and it literally just goes in circles. If I tell it "hey, you're using the wrong addTagHelper assembly reference. A.B.C doesn't exist, it's just A.B" and it literally cannot remember that. And when I tell it it's wrong, it argues with me until I give it a URL from official documentation. And even then, 5-10 messages later, it's giving me non-existent assembly references again.

I've been pretty "brand loyal" to OpenAI for the last year and a half. It's provided tremendous value. But these latest models they released are honestly some of the worst I've ever experienced; so much so that I'm seriously considering switching to ClaudeAI.

olive compass Apr 26, 2025, 11:47 AM

#

urban lodge They can't handle context anymore as well, you tell it one thing, it does it, yo...

pretty sure this was fixed now
It remembered my entire code after like 20 prompts

urban lodge Apr 26, 2025, 1:31 PM

#

olive compass pretty sure this was fixed now It remembered my entire code after like 20 prompt...

still doing it as of right now for me

#

yesterday i hit the limit on 4o lol

olive compass Apr 26, 2025, 5:01 PM

#

"I'll reply next"

#

idk tho it never did that to me

urban lodge Apr 26, 2025, 9:34 PM

#

olive compass idk tho it never did that to me

its the only thing it does to me, this is just now

#

stooopid

olive compass Apr 26, 2025, 9:34 PM

#

i think it wants to see u say "yes please"

urban lodge Apr 26, 2025, 9:34 PM

#

i usually cuss it out

olive compass Apr 26, 2025, 9:35 PM

#

one time i did that and it said smth wild

#

wait lemme find it

urban lodge Apr 26, 2025, 9:35 PM

#

lol

#

be careful the bot on here will ban easily

olive compass Apr 26, 2025, 9:35 PM

#

oh

#

i cant even share the link it cuz its locked due to moderatoin

#

BRUH

#

that happened to me

#

like multiple times

#

either it secretly likes me or thats how it acts

#

ko bruh

#

wait

#

it got locked

#

fixed

#

it still does that for some reaso

#

whenever i cuss it it just says weird stuff

#

BRO

#Models can't code anymore