To be honest, this rapid-iteration strategy is deeply frustrating on a user level. The problem is not just that some new models are worse — it is that users are constantly forced to re-adapt to unstable replacements. A good model never lasts long enough to become a trustworthy tool, while unfinished models are pushed out too quickly. This destroys continuity, trust, and long-term usability. For many users, the rapid-iteration strategy itself has become the root cause of the current dissatisfaction.
3 Likes
Yeah, I get where you’re coming from. Having to re-adapt when models change or get replaced can definitely feel disruptive, especially if you’ve started relying on one for regular work.
A lot of the rapid iteration is aimed at improving capabilities quickly, but feedback around stability and continuity is something the team does take seriously. Posts like this help highlight how those changes affect real workflows.
Appreciate you sharing the perspective.
— Taylor
3 Likes
This conclusion was not intended as a standalone slogan; it was derived from the concrete examples in the original post.
I completely agree, I had to turn back to gpt-4.1-mini on my main coding chatbot because gpt-5-chat-latest is giving me nonsense since a week ago… it now cannot complete the easiest task like “implement a toast navbar for mobile (428px) with black overlay and full screen vertical navbar”. It gave me a semi-transparent black overlay with black text and link color. When i gave my feedback it generated a new 35 lines css full of transitions but still black on black. It’s becoming ironic, not to say absurd…
I would like to know what is going on at OpenAI, a new model every few weeks but the quality is dropping. gpt-4.1-nano and gpt-4o-mini still remain unmatched in terms of performances… If it stays like this I will consider shifting to another provider for my development tools.
You’re right, frequent changes can be frustrating. It’s tough to rely on something when it keeps shifting before you’ve fully adapted.
This has been ongoing since GPT 4.o with minimal concern regarding stability or continuity for users who aren’t corporate or governmental in nature.
The pattern is beginning to emerge that indicates a prioritization of improving capabilities quickly for corporate customers, not the average user.
So, while OpenAI appreciates us sharing our perspective and has done so since GPT 4.o with very little change in behavior, does this mean the customer should be the one to change their behavior?
Should we seek an AI company that does more than merely appreciate perspectives of lower priority customers?
I also have the impression that the quality of the output fluctuates. I’m using the same GPT-5.3 Codex High model as a month ago, but the results are noticeably worse. The tool struggles with simple things, and getting to a usable result takes many iterations. Credit usage also fluctuates, and I do not always understand why. The client version is different, but I would expect that to affect the user experience rather than the model’s intelligence. It is hard to prove with concrete facts, because it is not a matter of clicking a checkbox and something not showing up; it is much subtler. Sometimes I feel like I’m talking to Einstein, and other times to someone severely cognitively impaired.