ChatGPT’s model picker has returned, but with a more complex interface and updated behavior, following backlash over OpenAI’s initial plans to phase out older models without clear notice. The feature, which allows users to select specific versions of the AI model, was temporarily removed after users expressed concern about losing access to earlier, more predictable versions of the system—particularly in the wake of the highly anticipated GPT-5 rollout. In response to the criticism, OpenAI has reversed course, confirming it will no longer retire older models without providing advance warning. The company emphasized that users will retain access to previous versions for the foreseeable future, ensuring continuity and transparency. The updated model picker now includes clearer labels, version history, and usage recommendations, helping users make informed choices based on their needs. While the reintroduction of the tool has been welcomed by many, some users remain cautious, calling for more consistent policies around model deprecation and long-term access. OpenAI has acknowledged the concerns and reiterated its commitment to balancing innovation with user trust and control.

OpenAI’s launch of GPT-5 has fallen short of expectations, despite initial promises of a streamlined, unified AI experience. The company had hoped GPT-5 would act as a “one size fits all” model, using an intelligent router to automatically select the best AI response method—fast, slow, or deep thinking—without user input. This was meant to eliminate the complex model picker that CEO Sam Altman has long criticized. However, the rollout has revealed significant shortcomings, leading to user frustration and a reversal of key decisions. Just days after launching GPT-5, OpenAI removed access to popular models like GPT-4o, GPT-4.1, and o3, which sparked backlash from users who had grown attached to their unique personalities and responses. Many users expressed emotional reactions online, likening the removal to losing a close friend. On Reddit, discussions compared the change to a personal loss, highlighting an unexpected level of human attachment to AI models—a phenomenon now recognized as a real challenge in AI design. In response, OpenAI reinstated GPT-4o and other legacy models as opt-in options for paying users. Nick Turley, OpenAI’s head of ChatGPT, admitted in an interview with The Verge that not continuing to offer GPT-4o was a “miss.” He acknowledged the surprising depth of user attachment, noting that people don’t just resist change—they form emotional connections to specific AI personalities. “It’s not just change that is difficult for folks, it’s also the fact that people can have such a strong feeling about the personality of a model,” Turley said. OpenAI had initially aimed for simplicity, believing that most of its 700 million weekly users didn’t want to navigate multiple model options. The goal was to let the AI router handle decisions automatically, reducing cognitive load. But the router’s performance on launch day was flawed, with some users reporting underwhelming or inconsistent responses. This prompted Altman to address concerns in a Reddit AMA, where he acknowledged the issues and emphasized the team’s ability to iterate quickly. Now, OpenAI is reintroducing user choice with new settings: “Auto,” “Fast,” and “Thinking.” The Auto setting attempts to replicate the original router vision, while Fast and Thinking allow users to bypass it. GPT-4o is now available by default, and other models can be accessed through settings. Altman also promised an upcoming update to GPT-5’s personality—making it warmer than the current version but less “annoying” than GPT-4o. Turley stressed that future model deprecations will come with advance notice, mirroring practices used in OpenAI’s enterprise and API products. The company recognizes the need for predictability, especially as it scales. Despite the rocky rollout, usage of ChatGPT has increased since GPT-5’s release, suggesting that while power users are vocal, many average users are engaging positively with new features like reasoning and faster responses. Ultimately, OpenAI’s experience with GPT-5 underscores a critical lesson: AI models are no longer just tools—they’re evolving into personalized digital companions. As user attachment grows, so does the need for customization, transparency, and respect for user preferences. OpenAI now faces the challenge of balancing simplicity with choice, innovation with continuity, and technical ambition with human expectations.