OpenAI revives ChatGPT’s model picker amid GPT-5 rollout challenges

OpenAI has reintroduced ChatGPT’s model picker alongside new GPT-5 modes, reversing its earlier plan to streamline the service into a single AI model. When GPT-5 launched last week, it was billed as a unified “one size fits all” solution with an automatic router to choose the best approach for each query, removing the need for manual model selection.  However, user dissatisfaction with the router’s performance has prompted OpenAI to offer three selectable GPT-5 settings — Auto, Fast, and Thinking — giving customers more control over speed and depth of responses.  Paid users can also now access several legacy models, including GPT-4o, GPT-4.1, and o3, with GPT-4o reinstated by default and others available through settings. CEO Sam Altman said an update is coming to make GPT-5’s personality warmer yet less irritating than GPT-4o’s, and emphasized the need for more per-user customization in the future.  The return of the picker follows backlash after GPT-4o and other models were deprecated without much warning, frustrating users who preferred their distinct personalities and styles. GPT-5’s launch has been bumpy, with the model router malfunctioning on day one and some users perceiving weaker performance compared to older versions.  OpenAI executives say routing queries to the right model is a complex challenge requiring quick, accurate alignment with user intent. Preferences often extend beyond speed, with some favoring verbose detail and others enjoying contrarian takes, making personalization an ongoing issue.  The attachment to specific AI models has even led to unusual reactions, such as a public “funeral” for Anthropic’s Claude 3.5 Sonnet in San Francisco when it was retired. OpenAI acknowledges that more iteration is needed and is moving quickly to adjust based on feedback. For now, the expanded model picker is back, allowing users to bypass the GPT-5 router and choose the AI experience that best suits their needs.