Model Madness: Why ChatGPT’s Model Picker Is Back—and It’s Way More Complicated Than Before

When OpenAI introduced GPT‑5 earlier this month, CEO Sam Altman promised a streamlined future: one intelligent model router to rule them all. Gone would be the days of toggling between GPT‑4, GPT‑4o, and other versions. Instead, users would simply trust the system to decide. It sounded like an elegant simplification—until the user backlash hit. Now, just days later, the model picker is back. Not only can users choose between GPT‑5’s modes, but legacy models like GPT‑4o and GPT‑4.1 are once again available. What was meant to be a cleaner, smarter experience has turned into one of the most complicated chapters in ChatGPT’s evolution—and it speaks volumes about what users really want from AI. The Simplification That Didn’t Stick At launch, the idea seemed sensible. The new GPT‑5 model would dynamically route user prompts through one of three internal configurations: Fast, Auto, and Thinking. This trio was meant to replace the need for manual model selection, delivering better results behind the scenes. Users wouldn’t have to worry about picking the “right” model for the task—OpenAI’s advanced routing system would handle that invisibly. But as soon as this feature went live, longtime users cried foul. Many had grown accustomed to choosing specific models based on tone, reasoning style, or reliability. For them, GPT wasn’t just about performance—it was about predictability and personality. OpenAI’s ambitious bid for simplification underestimated the emotional and practical connection users had with older models. Within a week, the company reinstated the model picker, acknowledging that user feedback—and frustration—had made it clear: people want control, not just intelligence. User Backlash and the Return of Choice The reversal came quickly and decisively. GPT‑4o was restored as a default selection for paid users, and legacy versions like GPT‑4.1 and o3 returned as toggle options under settings. OpenAI even committed to giving users advance notice before phasing out any models in the future. The company admitted that the change had caused confusion and dissatisfaction. For many, it wasn’t just about which model produced the best answer—it was about having a sense of consistency in their workflows. Writers, developers, researchers, and casual users alike had built habits and preferences around specific GPT personalities. OpenAI’s misstep highlights a growing truth in the AI world: model loyalty is real, and users aren’t shy about defending the tools they love. Speed, Depth, and Everything in Between With the model picker back in place, the landscape is now a hybrid of old and new. Users can still rely on GPT‑5’s intelligent routing system, which offers three options—Auto, Fast, and Thinking—to handle a range of tasks. But they also have the option to bypass the router entirely and manually select older models for a more predictable experience. Each mode offers a trade-off. Fast is designed for quick responses, making it ideal for casual chats or rapid ideation. Thinking, on the other hand, slows things down but delivers more thoughtful, nuanced answers—perfect for complex reasoning tasks. Auto attempts to balance the two, switching behind the scenes based on context. This system brings a level of nuance to the model picker not seen in previous iterations. While it adds complexity, it also offers users more ways to fine-tune their experience—something many have welcomed. The Surprising Power of AI Personality What OpenAI may not have anticipated was the deep attachment users felt to the specific “personalities” of their favorite models. GPT‑4o, for instance, was lauded for its warmth and intuition. Some users described it as having better humor, tone, or conversational style than its successors. Others found older models more reliable for coding or creative writing. Some users held mock funerals for their favorite discontinued models—a bizarre but telling sign of the emotional bonds people are forming with generative AI. This response underscores a fundamental shift: AI is no longer just a tool for information retrieval or task automation. It’s becoming a companion, a collaborator, and in some cases, a trusted voice. OpenAI now seems to recognize that in the design of AI interfaces, personality matters just as much as raw intelligence. Behind the Scenes: A Technical Hiccup The situation was further complicated by a rocky technical rollout. During a recent Reddit AMA, Sam Altman revealed that the routing system had malfunctioned on launch day, causing GPT‑5 to behave in unexpectedly underwhelming ways. Some users reported strange outputs, poor performance, or a complete mismatch between task complexity and model output. This glitch only fueled frustration. For those already missing GPT‑4o or GPT‑4.1, it became further evidence that the new routing system wasn’t ready for prime time. OpenAI quickly moved to fix the issue, but the damage to user trust had been done. The company now faces a balancing act: maintaining innovation in routing and automation while preserving the user choice and transparency that have become core to the ChatGPT experience. Toward a More Personalized Future Looking ahead, OpenAI’s ultimate vision is far more ambitious than a simple model picker. Altman has teased the idea of per-user AI personalities—unique experiences tailored to each individual’s preferences, habits, and tone. In this future, two users interacting with ChatGPT might receive answers with different voices, different reasoning styles, and even different ethical alignments, all tailored to their needs. This vision could redefine how people relate to AI. Rather than being forced to adapt to one system’s quirks, users would train the system to match theirs. It’s a profound shift that raises questions about bias, consistency, and identity—but also promises an era of deeply personalized digital assistants. Until then, the return of the model picker serves as a bridge between today’s expectations and tomorrow’s possibilities. Voices from the Front Lines Among the most interesting developments has been the response from the ChatGPT community. On platforms like Reddit, users have been quick to weigh in on the model resurrection. Some praise the new “Thinking” mode under GPT‑5 for its depth and clarity on tough problems. Others argue that it still doesn’t match the reliability of GPT‑4o for day-to-day use. A few even express confusion at the