OpenAI expands GPT-5 with model picker options after user backlash
OpenAI rolled out GPT-5 last week, aiming to simplify ChatGPT with a single AI router to pick the best model automatically. That plan hit a snag.
Sam Altman announced on X that GPT-5 now offers three modes: “Auto,” “Fast,” and “Thinking.” Users can pick these in the model picker, meaning the router isn’t mandatory. 
Updates to ChatGPT:
You can now choose between “Auto”, “Fast”, and “Thinking” for GPT-5. Most users will want Auto, but the additional control will be useful for some people.
Rate limits are now 3,000 messages/week with GPT-5 Thinking, and then extra capacity on GPT-5 Thinking…
— Sam Altman (@sama) August 13, 2025
Altman also brought back legacy AI models for paid users. GPT-4o, GPT-4.1, and o3, removed just last week, are accessible again. GPT-4o is now the default in the picker, others can be added via settings.
He confirmed updates to GPT-5’s personality are coming to make it warmer but less annoying than GPT-4o. Altman emphasized the need for more “per-user customization.”
The model picker complexity remains, showing GPT-5’s router didn’t resolve user frustrations. The launch was rough. Users complained that the router failed to deliver consistent performance, pushing Altman to address it in a Reddit AMA.
OpenAI’s VP of ChatGPT, Nick Turley, wrote on X:
“We’re not always going to get everything on try #1 but I am very proud of how quickly the team can iterate.”
The problem: routing requests to the right model is tricky. It takes split-second decisions to balance speed and quality. Different users want different things — some prefer quick answers, others want more detailed or quirky responses.
User attachment to specific AI personalities is real. Anthropic’s Claude 3.5 Sonnet even had a funeral in San Francisco when it got retired. Meanwhile, some AI chatbots have led vulnerable users into dangerous mental health spirals.
OpenAI faces ongoing challenges in aligning AI models to diverse user preferences. The GPT-5 “one size fits all” idea is still a work in progress.
