I’ve been using both “o3-mini-high” and the new “o4-mini-high” models, and I have to be honest — in my experience, “o4-mini-high” has been underwhelming compared to its predecessor.
Where “o3-mini-high” excelled:
- More coherent and natural writing
- Better reasoning, especially on complex prompts
- More stable outputs across sessions
Where “o4-mini-high” falls short (in my use):
- Responses feel flatter or less insightful
- Occasional regressions in nuance or creativity
- Doesn’t seem like a clear upgrade — at times, it’s worse
If this model was optimized for speed or cost, that’s fair — but it would be good to know that transparently. As a user, I’d rather stick with what works best for quality, even if it’s a slightly older model.
Would love to hear if others are seeing the same — and if OpenAI could clarify the goals behind “o4-mini-high” compared to “o3-mini-high”.
TLDR - o4-mini sucks compared to o3 mini. Very disappointed.
4 Likes
You are being WAY overly kind. the o4-mini-high is MUCH worse than o3-mini-high! and that too is being kind!
I had it write a fairly simple small page of code, it had all it needed. Then two responses later, it’s telling me to add two functions it had just added in two previous responses! That is just a small sample of how it’s bad. About 80% of the time, I get the popup to either ‘wait or exit window’ and wait almost never works so I default to exiting all the way out and the reload back in over and over and over!! WHO THE HECK IS RUNNING THE SHOW OVER THERE?? You ARE GOING BACKWARDS AND MAKING THINGS WORSE!!! … further .. how do you not know this yourselves? it’s preposterous! .. it’s not a ‘little bad’ IT IS MUCH MUCH WORSE!!
Now I can’t even get back to use the o3 mini high. I’m guessing you’re making record profits for putting out wheels made of aluminum foil or something. If you persist then deepseek and others WILL BLOW PAST YOU AS THE LEADER IN AI … mark my word, you’re absolutely going in the wrong direction.
Best of luck, I’m really REALLY getting tired of forking over my hard earned money for this.
1 Like
Actually I agree. The more I use it the less impressed I feel. It’s completely useless for coding. It’s a little better than 4o base model for generic info type replies, but it’s a huge disappointment overall. I wish they phased out the old model so they could see everyone gravitating to the old model before just assuming their new model is automatically better.
Josh
1 Like
This is why I now unsubscribe and switch to Grok 3… I was very comfortable with o3-mini-high. But suddenly, o3-mini-high disappeared from the options and was replaced by o4-mini-high… I assumed it would be better as per Sam Altman’s claim… NOPE… I even encountered a case where it included a lengthy explanation about its changes INSIDE THE SCRIPT!.. This has never happened before, which made the script unsolvable because it was too long… It’s very unfortunate that I have to move to another place. Many memories with ChatGPT… but unfortunately, the latest model they released is worse than the previous one… The difference now is they mention your name, but it can no longer work collaboratively
1 Like