ChatGPT model picker returns as GPT-5 rollout faces user feedback
ChatGPT’s GPT-5 rollout brings new modes and restores legacy models, reflecting OpenAI’s efforts to align AI performance and personality with user preferences.
OpenAI recently launched GPT-5, promising a simplified ChatGPT experience that would act as a unified AI model. The company intended GPT-5 to include a model router capable of deciding the best AI response for any user query automatically. OpenAI expected this approach to remove the need for a model picker, a menu of AI options that CEO Sam Altman has publicly criticised for being cumbersome.
Despite these expectations, GPT-5 has not fully delivered a unified solution. Altman announced that the model picker has returned, now offering new settings called Auto, Fast, and Thinking. The Auto setting works similarly to GPT-5’s router, while users can select fast or slow models directly to suit their preferences. This approach provides flexibility but also reintroduces the complexity OpenAI hoped to avoid.
What can users expect?
Paid users can also access several legacy AI models, including GPT-4o, GPT-4.1, and o3. GPT-4o is now available in the model picker by default, while the other models can be added through ChatGPT’s settings. Altman highlighted ongoing improvements to GPT-5’s personality, aiming for a warmer interaction style that avoids the quirks of previous models. He also emphasised the importance of offering users more personalised customisation of AI personalities in the future.
The rollout of GPT-5 has been met with mixed reactions. The temporary deprecation of GPT-4o and other models sparked backlash from users attached to their favourite AI responses. Some users were disappointed with the router’s performance on launch day, prompting Altman to address concerns in a Reddit AMA session. OpenAI acknowledges that routing prompts to the correct AI model is a complex task. The system must quickly align responses to both user preferences and the specific query while maintaining fast response times.
Human attachment to specific AI models adds another layer of complexity. Users may prefer certain models for their verbosity, tone, or ability to provide contrarian responses. In some cases, people have formed emotional connections to AI personalities, as seen when hundreds attended a funeral for Anthropic’s Claude 3.5 Sonnet model. Such phenomena illustrate the growing impact of AI on human behaviour and the need for careful design and alignment.
OpenAI continues to iterate on GPT-5 and the model picker to balance performance, speed, and personality customisation. Nick Turley, OpenAI’s VP of ChatGPT, expressed pride in the team’s rapid improvements and emphasised that the company expects to refine the experience over time. The addition of multiple model options ensures that users can select AI interactions that meet their expectations, preserving familiarity while exploring new capabilities.
GPT-5 and the model picker highlight the challenges of delivering a one-size-fits-all AI experience. OpenAI’s work reflects both technological hurdles and evolving user behaviour. As the company continues refining GPT-5, the goal remains to offer AI models that adapt intelligently to individual preferences while remaining fast, responsive, and engaging.
