About 10 days ago I posted here because I felt GPT-5.5 had become noticeably worse for software development. Interestingly, over the following days the behavior seemed to improve again. Coding felt more focused, changes were more localized, and overall reliability was closer to what I had originally experienced.
However, today something feels very different.
Throughout the day I’ve repeatedly received messages saying GPT-5.5 is overloaded and suggesting that I switch to another model. At the same time, coding performance feels dramatically worse again.
The issues I’m seeing are the same ones that originally brought me to this thread:
-
making changes outside the requested scope;
-
touching files that were never mentioned;
-
introducing hidden modifications that I only discover during review;
-
wandering around the repository instead of staying focused on the task at hand;
-
struggling with tasks that would have been trivial before.
I want to be careful here because I don’t have visibility into OpenAI’s infrastructure and I don’t know what’s happening behind the scenes.
But as a paying user, it genuinely feels like coding quality fluctuates together with model availability and load.
Maybe it’s routing. Maybe it’s capacity constraints. Maybe it’s something else entirely.
All I know is that the experience I had today is very different from the experience I had last week.
I’m paying for GPT-5.5 specifically because of the quality level I experienced when it was working well. If the model is overloaded, I’d much rather wait in a queue than silently end up with an experience that feels materially different from what I’m paying for.
Is anyone else seeing the same pattern today?