OpenAI has reintroduced a model picker for ChatGPT following user feedback on the initial rollout of its new GPT-5 model.
The company’s initial strategy aimed to simplify the user experience by using a single, unified AI model capable of automatically selecting the best response for any query. However, this approach, which was intended to replace the model picker, was met with mixed reactions.
OpenAI CEO Sam Altman confirmed the return of the model picker, which now includes new settings: Auto, Fast, and Thinking. The Auto setting functions similarly to the GPT-5 router, while the other two options allow users to manually select between faster or more thoughtful models, providing greater flexibility.
User Feedback and Legacy Models
The temporary removal of legacy models, such as GPT-4o, during the initial launch drew criticism from users who had developed a preference for specific AI personalities and response styles. To address this, OpenAI has made these models available to paid subscribers. Users can now access GPT-4o by default and add other legacy models, including GPT-4.1 and o3, through the settings.
Altman has also emphasized ongoing efforts to refine GPT-5’s personality, aiming for a “warmer” and less idiosyncratic interaction style. The company recognizes the growing emotional connections users form with AI models, a phenomenon highlighted by the widely reported “funeral” for Anthropic’s Claude 3.5 Sonnet. This underscores the complexity of balancing technological performance with user expectations and personal preferences.
Challenges and Future Outlook
The reintroduction of the model picker highlights the significant challenges in creating a “one-size-fits-all” AI solution. Routing prompts effectively requires the system to quickly align responses with user preferences, query specifics, and performance requirements. Nick Turley, OpenAI’s VP of ChatGPT, noted the team’s rapid improvements and expressed confidence that the experience will continue to be refined over time. The company’s focus remains on offering intelligent, responsive AI models that can also be customized to individual user needs.
