This week's launches: o3, o4-mini, GPT-4.1, and Codex CLI

bruh, o4 mini and o4 mini high didnt follow instruction
so many errors

2 Likes

How do i verify via azure? you certainly can’t expect us to update our billing information via openai

I can’t believe I don’t have access to model o1 anymore. The new models hallucinate and fail in following instructions while o1 had a great success ration. This is a significative step back.

5 Likes

O3 - Our most powerful model that constant errors out when you give it more than two pages of text.

5 Likes

There is no reasoning at all, only constant errors. For a small output, I have to spend hours making corrections. It is impossible to work with the new models. If o1 does not return, or something similar, we will cancel our Team subscription. It is too time consuming and not possible to work with. Not serious!

5 Likes

Yeah, all my previous projects and chats are dead and not working with new models - total mess, unfortunately.

3 Likes

As a ChatGPT Pro user, I totally agree with this man.
o3’s accuracy is far lower than o1’s and it keeps repeating detailed errors, so the overall quality has dropped significantly—this is clearly a downgrade.
o3 feels less like a reasoning model and more like a hallucinating fiction writer. It repeatedly loses track of context and invents details that were never provided, even worse than 4o.

8 Likes

All new models are complete crap.

With the PLUS or PRO plan, any old user, when using the current one, will be tremendously disappointed and cancel the subscription.

NOTE: This makes me very, very sad. A new reality.

4 Likes

I don’t understand why none of the chatgpt staff, including the guy who posted about the new models, don’t give any official information to any customer. Does the company consider itself so superior to its own customers/users?

If chatgpt became what it is, it’s because of its customers and not just because of its own merit.

I think it’s also a matter of respect. You simply stop offering an excellent service for a new one that isn’t usable, and you don’t even care to the point of not giving any communication or information. That’s ridiculous.

3 Likes

In recent years, I have been a ChatGPT user: I started with model o1 and then upgraded to o1-pro. In my experience, o1 delivered more consistent and faster results than o3, and I’m glad to know I’m not the only one who has encountered difficulties programming with this version.

After evaluating its performance, I have decided to cancel my PRO subscription. My only remaining benefit is access to o1-pro, so I will now switch to Gemini. I hope the company adjusts its course, as the transition from DeepSeek and Google to o3 has fallen short of expectations.

3 Likes

Thanks for the information, I did the same thing and had success. Thanks for the advice =)

You guys are becoming increasingly disingenuous with everything you do and you’re going to pay for it dearly.

Stop telling us your models are godlike coders when the Claude and Gemini models DESTROY whatever garbage you’re serving up in the app or via API.

4.1 was such a waste of time I think you should pay society damages for the resources you wasted creating these things that are worse than Claude and Gemini’s models two generations ago.

What happened? You must realize most of us who pay are not from 3rs world countries who can’t afford to pay thousands per month if necessary.

Just make it pay to play in the app and drop these outdated Web2 predatory pricing models. Evolve.

Every developer that uses your models would prefer a true coding only model. Drop the multi-modal fluff and give us a model that is worthy of the coding Eli scores you proudly present to us that aren’t real or available to us in any form.

Stop trying to be budget about this as coders cost $10k or more per month. I’d gladly pay $1000 a $2000 per month if you give me a flawless senior level coding engineer to work with as would most everyone in the west who runs a business.

Know your audience and stop pushing trash. Claude-sonnet 3.7 thinking has it right as I can budget as many thinking tokens as I want and it just crushes any code I throw at it.

3 Likes

What a regression, OpenAI! The launch of o3 and o4 feels like a step backwards. They’re terrible. I can’t continue the tasks I had been working on with o1.
The only option left is using 4o, which struggles with even moderately complex tasks, crashes constantly, and suffers from context token limitations.
I have projects on hold because you removed o1 from the Plus plan with no prior notice.
It’s impossible to afford the Pro plan in my country — it costs as much as a minimum wage.
So the question is: why should I keep paying for a service that’s worse than the free one?

4 Likes

I agree 100% with all the negative feedback posted here. After multiple years of heavy daily use, after experiencing the breakthrough that was o3-mini-high, you’ve deprecated it and replaced it with impaired models that talk themselves in circles before running out of tokens and spit out worthless results.

You’ve just lost another Pro subscriber. I hope you lose many more… you deserve it.

Gemini 2.5 Pro to the rescue.

4 Likes

@edwinarbus - 4.1 shows amazing results - BUT! - File Search or RAG is completely useless… any outlook for correcting this bug??

2 posts were merged into an existing topic: O3 and o4-mini are extremly bad lazy and not suitable for coding anymore

Could you be a bit more specific? all the 4.1 variants?

I personally find 4.1 fine for tools and RAG, but 4.1 mini and nano definitely have issues distinguishing between two subtly different tools (something 4o-mini does not suffer from)

1 Like

Thank so much for the updates. For a couple of days my 4.1 assistant has been busted. No more connection to the vector store. Something bad must always happen.

This is a downgrade, gpt 4 turbo could handle scripts of a lot more lanes without any issue.

4.1 is not even comparable to it.

i’m writing it now, cause i got the upgrade yesterday where i live, this upgrade totally destroyed my workspace.

1 Like