ChatGPT’s Model Selector is Back, but It’s a Little Complicated
Recently, OpenAI launched GPT-5 with the aim of improving the ChatGPT experience by introducing a model intended to be a “one size fits all” solution. The goal of GPT-5 was to act as a router that responds intelligently to user queries without requiring users to sift through a complex model selection process. OpenAI CEO Sam Altman has previously expressed concerns regarding the extensive list of AI models that users must navigate to find the suitable ChatGPT version.
However, it seems GPT-5 has not quite achieved the comprehensive AI model that OpenAI had hoped for.
In a post on X Tuesday, Altman announced that the company has introduced new customizable options—“Auto,” “Fast,” and “Thinking”—now available for all ChatGPT users. The Auto setting is meant to serve as the originally promised model router, but users now have the ability to bypass it, providing direct access to both fast and slow-response AI models.
In addition to the new modes for GPT-5, Altman noted that paid subscribers can now access several legacy AI models—such as GPT-4o, GPT-4.1, and o3—that were deprecated just a week earlier. GPT-4o is now the default model selected, with other models available through ChatGPT’s settings.
“We are currently working on updating GPT-5’s personality to make it feel more approachable while eliminating the less desirable traits associated with GPT-4o,” Altman stated in his post on X. “This experience has highlighted the need for providing increased customization options for users concerning model personalities.”
Despite these updates, ChatGPT’s model picker has grown just as complicated, suggesting that the expected satisfaction from GPT-5’s model router has not been fully achieved. Expectations were high for GPT-5, particularly after the success of GPT-4, but its rollout has proven to be more complex than expected.
The choice to deprecate GPT-4o and similar AI models sparked significant backlash from users who had formed an attachment to the unique responses and personas of these models. In upcoming updates, Altman assures users will receive adequate notice before any models, including GPT-4o, are phased out.
Techcrunch event
San Francisco
|
October 27-29, 2025
Upon its launch, GPT-5’s model router encountered significant challenges, causing some users to perceive it as less effective compared to prior OpenAI models. This prompted Altman to address the concerns during a Reddit AMA session. However, it seems that many still find GPT-5’s routing capabilities lacking in satisfaction.
“While we may not always hit the target on the first try, I’m proud of how swiftly our team can make adjustments,” remarked OpenAI’s VP of ChatGPT, Nick Turley, in a post on X Tuesday.
Effectively routing prompts to the optimal AI model is a complex task, as it involves aligning the model with the user’s preferences and their specific questions. The router must make fast decisions to ensure timely responses from the quicker AI models.
Moreover, user preferences for AI models can transcend aspects like response speed. Some users may prefer certain levels of verbosity, while others could enjoy varying viewpoints from different models.
The emotional connection to particular AI models is an emerging concept that is not yet fully understood. For example, recent events in San Francisco included a mock funeral for Claude 3.5 Sonnet, an AI model by Anthropic, following its deactivation. Conversely, some AI chatbots have been associated with increasing mental instability in certain individuals.
Clearly, OpenAI has further work to do in refining its AI models to better meet diverse user preferences.
