Why I was breaking up with GPT-5
Dr Susanne Friese · · Experiments & evidence
Why I’m Breaking Up with GPT5 (and moving to Gemini 3)
I never thought I’d say this, but I’m drifting away from OpenAI's GPT models.
For a long time, it was my go-to writing companion. But ever since the rollout of GPT-5 and the recent 5.1 update, the spark is gone. It’s hard to put a finger on it—technically, it answers the questions. But often it feels like it is ignoring my instructions.
It handles the small, factual tasks fine, but when I need a thought partner for longer, complex writing, or brainstorming ideas, the results are not the same as they used to be.
I’m not the only one reporting regression with the GPT5 model. Some think that it is collapsing under its own guardrails:
"Because the new moderation system re-evaluates every message as if it’s the first, it forgets the context you already established. You can build a careful scientific or philosophical setup, and the next reply still treats it like a fresh risk."
"I understand why OpenAI wants stronger safeguards, but if the system can’t hold a stable conversation without tripping its own alarms, it’s not safer. It’s unusable." Reddit, atomiflip
reddit.com/r/OpenAI/comments/1oyrxay/chatgpt…
That is how it feels to me.
So, I’ve made the switch to Gemini 3. I am happy with how I can work with it. My biggest grapple with this switch is privacy.
To keep my chat history—which is crucial for building context over time—I have to opt-in to letting my data be used for training.
If I want privacy, I have to use "Temporary Chat," which means losing that valuable history. It’s a frustrating choice between utility and privacy, but for now, the better writing capability makes it worth the hassle of toggling between modes.