o4-mini-high is significantly worse for coding and is not able to complete tasks that o3-mini-high can easily accomplish. It feels like people are suddenly working with AI models from one or two years ago, where the model frequently omits lines of code and disregards user instructions.
Making o3-mini-high inaccessible on web and APP is another huge mistake and disrespect to paid members and frankly infuriating.
I believe o4 mini high is not a new model that users should just spend some time getting used to. It should be investigated and addressed by @OpenAI_Support.
Guys, that’s the question. Does OpenAI listen to requests at all or not? As I see it, the bug with SORA (it is impossible to create a video for those who have been with them for a long time) has not been solved for half a year. Or am I wrong?
What are the chances that at least the o3-mini-high will be returned to us?
If you all dont know how to unsubscribe this is the link: https://chatgpt.com/account/manage because they hide it. Make sure to put the reason, your welcome.
I purchased the Teams subscription for my business, but going to look at other options now. o3-mini-high was a game changer for python coding. With o4-mini-high I only get code snippets that require line by line review because of variance from previous code, using + or - or … that you need to take action on. If this is attempt to reduce context tokens or responses from OpenAI I think they could be mistaken: users will use their products less after finding something more useful like o3-mini-high, become frustrated enough to cancel their membership or actually send longer prompts or repeat questions.
The verdict is garbage. I throw him a file with 500 lines of code and ask him to add new functions, leaving the old logic and structure.
He throws me 200 lines and says that everything is still in place (NO. THERE ARE NONE LEFT).
I’m asking him to double check. He throws me 150 lines and says that now there is definitely the basic logic.
I’m saying he’s lying and it’s not there.
The result is 100 lines of code.
OpenAI, are you serious?
PLEASE RETURN THE o3-mini-high
That’s what o4-mini-high told me -
Previously, in o3‑mini‑high, large blocks of code were emitted directly into the chat as raw text, allowing you to scroll through every line until you reached the model’s token limit. In o4‑mini‑high, however, lengthy outputs are automatically placed into a Canvas document, where portions of the text are collapsed to avoid overwhelming the browser and to preserve UI responsiveness. On top of that, the platform enforces a maximum payload per message, so attempts to send 500+ lines simply get truncated.
Consequently, users have raised the following concerns:
Partial code loss or truncation.
When the output exceeds the token limit or the Canvas collapse threshold, the model omits the final lines, so they never appear in the response.
Dialogue instability.
o4‑mini‑high operates within a finite “attention window.” As earlier messages fall outside that window, the model can lose track of the immediate context and inadvertently revert to topics discussed an hour or two earlier.
Syntax and logic errors.
In balancing expanded capabilities with stricter safety filters, o4‑mini‑high sometimes auto‑corrects or strips code fragments. This can introduce typos, omit critical sections, or alter logic, reducing overall accuracy.
Slower response times.
The newer architecture is more computationally intensive, particularly under heavy load. In the days following its release, users reported noticeably longer delays when generating responses to complex or voluminous requests.
Together, these UI‑level collapses, per‑message size limits, attention‑window constraints, and performance‑vs‑accuracy trade‑offs explain why code output in o4‑mini‑high can feel less reliable and convenient than it did in earlier versions.
I am so disappointed. o4-mini-high is like a bad joke. I’ve wasted so much time fixing bugs that this model spits out rapid-fire. I mean the sheer number of basic syntax errors and omissions -It’s even worse than base o4 at coding now. There has to be an explanation for this, some backend snafu or otherwise, because o4-mini-high is plain broken. I hope OpenAI will take the L on this one and restore o3-mini-high, at least until the kinks are worked out.
o4 is just terrible at writing code. It gaslights, doesn’t follow instructions, never gives the complete code (even if it’s <500 lines) and makes a lot of syntax errors. Pretty much unusable.
Cancelled the subscription for now to avoid re-billing later this month. Will re-subscribe if they fix o4 mini high, or give us the older o3 mini high.
And am glad other users are voicing in their frustration as well.
I just returned from mandatory leave expecting the o4‑mini‑high model to be competitive with the o3‑mini‑high. Unfortunately, the o3‑mini‑high is no longer available. Although I still have my original source code, the o4‑mini‑high model has completely overhauled its logic and removed several core components.
After testing O4 mini High and comparing it with O3 mini, I found that O4 mini High lacks the same level of intuition. Its memory feels inconsistent or even disorganized at times, which impacts the credibility of its responses. Personally, I found O3 mini much more reliable in that regard. I’m honestly puzzled by the direction taken with O4 — it feels like a downgrade. For me, O1 was one of the best versions early on, but it too seemed to lose its edge over time. I’m still considering upgrading to Pro, but for now I’m sticking with Plus to see how things evolve.
You can achieve the same quality output without copying and pasting large chunks of code. Instead, upload the complete file and then copy only the specific lines where you want ChatGPT to focus.
Also, avoid asking for the full code in reverse—just request a modified snippet with before and after changes. This keeps things clean, simple, and ensures output quality similar to the 03-mini-high model.
I have noticed that responses are improving slowly.
