When OpenAI dropped GPT‑5 last week, the pitch was crystal:
One model to rule them all. That means: No more fiddling with the model picker. No more wondering, “Do I want the fast one, the smart one, or the moody one?” Just one brain with a built-in router that picks the best approach for your prompt.
Except… plot twist — the model picker is back. And it’s busier than ever.
Sam Altman hopped on X Tuesday to announce GPT-5 now comes with three selectable modes, including:
Auto — the original router idea, that picks for you.
Fast — quicker responses, less deep thinking.
Thinking — slower and more detailed answers.
Oh, and remember how GPT‑4o, GPT‑4.1, and o3 got axed last week?
Well, they’re back — but only for paid users. GPT‑4o is in the picker by default; the rest you can toggle in Settings.
But in case you missed it, launch week was… messy:
Day one router glitches made GPT-5 feel less capable than the “oldies.
The “simplify everything” dream… ended up making the picker more complex.
And the kicker? People didn’t just want the “best” model — they wanted their model.
Now that’s the part OpenAI didn’t see coming.
Some folks love the verbosity of one model. Others like the sass of another. And then you’ve got people in San Francisco holding an actual “funeral” for Anthropic’s Claude 3.5 Sonnet when it was retired.
In OpenAI’s case, it was a coordinated revolt that had the company caving within hours — and now its retired, much-loved older models have been resurrected like fan-favorite TV characters.
But that’s not all:
Altman also teased a “warmer” GPT‑5 personality coming soon — something friendlier than the current vibe, but without GPT‑4o’s over‑eager cheerleader energy.
The Long‑term goal? per‑user personality customization so ChatGPT doesn’t just answer the way you want — it acts the way you want.
Bottom line:
OpenAI’s real challenge isn’t building the smartest router or model — it’s figuring out how to upgrade AI without breaking people’s weird, emotional bond with their favorite bots.
Read the full report.
