When OpenAI announced GPT-5 last week, I thought finally, one AI model to rule them all. The company promised a simpler, smarter ChatGPT experience with GPT-5 acting like a router. You’d ask a question, and the system would automatically pick the best model to answer — fast, easy, frictionless. That was the idea.
But here we are, and the model picker is back. And honestly? It’s just as overwhelming as before.
What’s New: Auto, Fast, and Thinking Settings
On August 13, OpenAI CEO Sam Altman dropped the news on X (formerly Twitter). Now, when you’re in ChatGPT with GPT-5, you can choose between three settings:
- Auto: This uses the new routing system. Think of it as letting the AI pick what kind of brain it should use behind the scenes.
- Fast: Prioritizes speed in responses.
- Thinking: Slower, but designed for more thoughtful or complex replies.
In theory, Auto is supposed to be good enough for most users. But the fact that they had to bring back manual choices? That says Auto isn’t winning everyone over.
Why Users Are Frustrated
This whole thing started with good intentions. OpenAI wanted to retire the confusing model menu — something Altman himself admitted he hated. But after rolling out GPT-5 and removing older models like GPT-4o and others, users weren’t happy. People had grown attached to the older AIs — their quirks, their tone, their reliability. And they weren’t ready to let go.
It didn’t help that GPT-5’s router seemed broken on launch day. Some people felt the model wasn’t as sharp or responsive. Others missed the specific “vibe” of GPT-4o. It got messy fast.
So now, OpenAI backtracked a bit.
According to Altman, paid ChatGPT users can once again access several legacy models, including GPT-4o, GPT-4.1, and o3. GPT-4o is even back by default, with the rest available in settings.
So What’s the Deal with GPT-5?
GPT-5 isn’t what people expected. OpenAI hyped it up as the model that would eliminate the need to pick between different brains. Instead, it’s kind of become a new version of the same problem — just with fancier labels.
Nick Turley, VP of ChatGPT, admitted as much. “We’re not always going to get everything on try #1,” he posted. But he praised the team for how quickly they respond to issues. It’s an honest take, and one that acknowledges how tough it is to route prompts automatically — especially when people have wildly different preferences.
Some like their AI fast and brief. Others want deep and detailed. And some just miss the tone and personality of a model they’d gotten used to.
Altman did mention that an update to GPT-5’s personality is coming soon — aiming for something “warmer” than the current version, but “not as annoying (to most users)” as GPT-4o. That’s not exactly glowing praise, but it’s a hint they’re listening.
The Big Picture
This whole rollout shows just how personal AI tools have become. People don’t just use them — they form opinions, habits, even attachments. When OpenAI retired models, it underestimated just how strongly some users felt about them.
And while GPT-5 is powerful, it clearly hasn’t solved the experience puzzle yet. Giving users more control is a start. Fine-tuning the personality side of GPT-5 might help too.
But the bigger challenge? Building AI that aligns to what each person wants — not just in terms of speed or accuracy, but in how it feels to interact with.
OpenAI’s working on it. But for now, if you open up ChatGPT and feel like the interface has déjà vu… you’re not alone. We’re still picking models, tweaking settings, and figuring out what kind of AI experience we really want.
