Remember when you could choose your favorite model in ChatGPT? That option disappeared with the launch of GPT-5, sparking a wave of frustration from users who missed the specific feel and performance of models like GPT-4o. Well, after a swift and intense backlash, OpenAI has brought the model selector back—but the new system is far more complex than its predecessor.
When GPT-5 was first released, OpenAI’s goal was to simplify the user experience by using a real-time router to automatically select the best version of the new model for each task. The model picker, which previously allowed users to manually switch between options like GPT-4o and GPT-3.5, was removed. This change, however, was met with strong backlash. Many users expressed that the new GPT-5 model felt “colder” or “dumber” and lacked the specific personality and performance they had come to rely on from GPT-4o.
The Return of Choice, with a Catch
In response to the feedback, OpenAI CEO Sam Altman admitted that the router system was “underperforming” and quickly reversed course. The model picker has now returned, but its functionality varies significantly based on a user’s subscription tier.
- Paid Subscribers: Plus, Team, and Enterprise users now have a “Show Legacy Models” option in their settings. This allows them to switch back to models like GPT-4o, GPT-4.1, and even GPT-4.5, catering to users who were unhappy with the change. They can also manually select different modes of GPT-5, such as “Auto,” “Fast,” and “Thinking,” to better suit their needs.
- Free Users: For now, users on the free tier remain limited to the new default GPT-5 model and do not have access to the legacy models.
This “complicated” new system highlights the tension between a developer’s desire for a streamlined, single-model experience and a user base that has grown attached to the specific characteristics of older models. While OpenAI has promised to improve GPT-5’s personality, the company’s decision to reintroduce older models for paid users acknowledges that a one-size-fits-all approach is not yet a viable solution for the diverse needs of its user base.
