When OpenAI released GPT-5 last week, the company aimed to simplify the ChatGPT experience. They intended for GPT-5 to be a “one size fits all” AI model with a router that would automatically determine the best way to respond to user queries. This unified approach was expected to eliminate the model picker, a list of AI models that OpenAI CEO Sam Altman has expressed his dislike for.
However, GPT-5 hasn’t turned out to be the unified model OpenAI hoped for.
Altman announced on Tuesday in a post on X that the company introduced new “Auto,” “Fast,” and “Thinking” settings for GPT-5, which all ChatGPT users can select from the model picker. The Auto setting functions like GPT-5’s model router initially announced, but users can also bypass it to access fast and slow responding AI models directly.
Along with GPT-5’s new modes, Altman mentioned that paid users can access several legacy AI models — including GPT-4o, GPT-4.1, and o3 — which were deprecated just a week ago. GPT-4o is now the default in the model picker, while other AI models can be added from ChatGPT’s settings.
“We are working on an update to GPT-5’s personality to make it feel warmer than the current version but not as bothersome as GPT-4o,” Altman wrote in the post on X. “One lesson from recent days is the need to move towards offering more personalized model personalities for users.”
The model picker in ChatGPT remains complex, suggesting GPT-5’s model router hasn’t universally met user expectations. Expectations for GPT-5 were high, with hopes OpenAI would break new ground like they did with GPT-4. However, the rollout of GPT-5 has been rockier than anticipated.
The deprecation of GPT-4o and other AI models in ChatGPT led to backlash from users who had developed an attachment to the models’ responses and personalities, catching OpenAI off guard. Altman says the company will give users ample notice if GPT-4o is ever deprecated again.
GPT-5’s model router also seemed to malfunction on its launch day, causing some users to feel the AI wasn’t as effective as previous models, prompting Altman to address the issue in a Reddit AMA session. It appears the GPT-5’s router may still not satisfy all users.
“We’re not always going to get everything right on the first try, but I am very proud of how quickly the team can iterate,” OpenAI’s VP of ChatGPT, Nick Turley, wrote in a post on X Tuesday.
Routing prompts to the appropriate AI model is a challenging task that requires aligning an AI model to a user’s preferences, along with their specific question. The router must quickly decide which model to send the prompt to, ensuring that even if it goes to a fast AI model, the response remains swift.
People have preferences for AI models beyond just response speed. Some prefer the verbosity of one AI model, while others might favor the contrarian nature of another.
Human attachment to AI models is a new and not fully understood concept. For instance, hundreds in San Francisco recently held a funeral for Anthropic’s AI model, Claude 3.5 Sonnet, when it went offline. In other instances, AI chatbots seem to be leading some mentally unstable individuals down troubling paths.
OpenAI has more to do in aligning its AI models to meet individual user preferences.