I use the ChatGPT Mac app for work with coding. I give it 1,000 line files MAX and ask it for help like finding duplicated logic flows, or I’ll give it a test file stub (like mocked dependencies already created) and a copy of a service and ask it to make tests for me.
o3-mini-high was great at this, I used it all the time. I used o1-pro sometimes but it wasn’t as good for code in / code out tasks, I used it more for ideas around scoping at how I’d organize new code, stuff like that.
o4-mini-high nor o3 follow instructions and are basically unusable for what I used to use it for. I’m not sure if o1-pro is the same but it seems like it’s been nerfed.
I was ok w/paying $200/mo b/c it saved me a lot of time, it wasn’t always perfect but generally I could fly through its output and use it as a speed up.
Over the last few days it’s been a giant waste of time, repeating the wrong things and not following instructions.
I have no idea what’s going on, I reached out to support and they haven’t replied.
I wish they’d bring back o3-mini-high and restore o1-pro to whatever it was before this week. This sucks.
Since the recent update, it’s become a complete mess. It’s impossible to work with ChatGPT anymore.
Projects with just 30 files (and low complexity) — ChatGPT can no longer send complete code snippets over 500 lines. It always sends poorly done fragments.
The company keeps getting worse. You might as well go to DeepSeek and code with it — you’ll get BETTER results than with Mini High.
At first, I even thought my account was limited, but that’s not the case — the company intentionally restricted developer usage to reduce resource consumption.
It’s a way to cut costs — if you use too many resources, you get penalized.
Just like a casino — the house always wins. Same with GPT. If you use a lot, well… too bad!
Use the API and pay for tokens — or screw yourself.
What’s most frustrating for me is that the projects I built with ChatGPT-4 (back when there was no ChatGPT O3 High Mini/ 4 o Mini , 4.5, or any of these current versions) are now practically impossible to maintain.
It takes 10 back-and-forths with 4Mini and it still doesn’t work—every fix is wrong. Even after creating multiple new chats to start from scratch, minimizing token usage from history, trying every way possible to explain things clearly—still, it’s impossible.
Back then, ChatGPT-4 could generate code, even if it had to send it in multiple parts. Now it just can’t.
It’s shameful.
Can you recommend any alternatives?
Maybe DeepSeek via API? (Their cost is WAY [around 10x] cheaper than ChatGPT.)
That way, by paying per request, I might be able to maintain the projects I originally built with ChatGPT.
It’s frustrating. Exhausting.
How is it possible? Losing hours on things that used to take minutes.
I think they are aware of this right now . I expect it was too expensive … now they gimped it … ppl are not satisfied .. which probably they will come up with a new model and a new pricing strategy
100% - ever since the update, it is basically useless for debugging, and code generation, even with simple issues. Super unhappy with the new oX models. Hope they fix this soon.
100% agree. GPT has gone backwards. I’ve even tried their GPT 4.1 and although they claim it’s great for coding due to greater token limit, it’s far worse than o1 and even o3 mini high. OpenAI’s become a scam.
Yeah, my friends… it’s complicated. Honestly, out of all the updates so far, this has been the worst one ever — and the worst performance I’ve seen.
I’d really like to know how many more people are going through this, because it’s impossible to tell if it’s a model-wide issue or something that’s just affecting us specifically.
i just bought the pro mode hoping for a better models than o3-mini-high now the answers feels like gpt 3.5 whats going on this is a scam im paying 200 dollars because i believed that this subscribtion will give me access to models that will reduce a loooot of time for me the newer models cant even handle beyond 500 lines of codes it became completly useless im never paying again
I really don’t know how they get away with this piss poor AI. It has never met a deadline, always asks for the same exact amount of time it just asked me for previously, completely violates repeated instructions, and losses a bunch of information. It has become more of a waste or time, a hindrance more than help. Which models work best for you because I tired all and only 4.5 worked well but I was only able to use it for 4-5 questions.
You are not kidding. o3-mini-high was amazing a week ago. I think they phased it out on 4/16/2025. I canceled my $200 subscription. It’s virtually worthless now.