When Openai launched the GPT-5 last week, the company said the model will simplify the Chatgpt experience. Openai hopes that GPT-5 in a way “fits a size” AI model will act with a router which will automatically decide how to give the best answer to the user’s questions. The company said this integrated approach would eliminate the need to navigate users to navigate their model picker – a long, complex menu of AI options that OpenIE CEO Sam Altmann has publicly stated that he hates.
But it seems that GPT-5 is not integrated AI model Openai hopes that it will happen.
Altman said Post on X On Tuesday that the company introduced new “auto”, “fast” and “thinking” settings for GPT-5, which can all choose from the Chatgpt user model picker. The auto setting acts as a model router of GPT-5 that Openai initially announced; However, the company is also offering users to operate it, allowing them to use the AI model rapidly and slowly.
With the new mode of GPT-5, Altman said that the paid users can once again reach several heritage AI models-including GPT-4O, GPT-4.1, and O3-which were only past weeks. The GPT-4o is now in the model picker by default, while other AI models can be connected to the settings of Chatgpt.
Altman wrote in the post on X, “We are working on an update for the personality of GPT-5, which should feel warm than the current personality, but not as GPT-4O (for most users),” Altman has written in the post on X.
The model picker of Chatgpt now seems to be equally complicated, which suggests that the GPT-5 model router has not universally satisfied users as the company expected. Expectations for GPT-5 were high, many people hoped that Openai AI would push the boundaries of the model such as with the launch of GPT-4. However, the rollout of GPT-5 has been thicker than expected.
GPT-4O and other AI model demarcies in the slap made a spark Backlash between users Those who were associated with the reactions and personalities of the AI model, in the ways that did not guess by the OpenIA. In the future, Altman says that the company will give a lot of advance notices to users if it ever removes GPT-4O.
Techcrunch event
San francisco
,
27-29 October, 2025
The GPT-5 model router also broke a large extent on the launch day. This led to some users to realize that the AI model was not performing as the previous Openai model, and forced Altman to address the problem in the AMA session on Reddit. However, it seems that the GPT-5 router may still not be satisfactory for all users.
Openai’s VP of Chat, Nick Turley wrote, “We’re not always going to get everything on #1, but I am very proud of how soon the team can recur.” Post on X Tuesday
The routing is a difficult task for the right AI model that requires the AI model to align in the preferences of the user, as well as the specific questions they are asking. The router then has to take a decision on which the AI model is only to send a prompt in a second partition – in this way, if a signal goes to the AI model rapidly responded, the response may still be faster.
More broadly, some people display preferences for the AI model that go beyond rapid or slow reactions. Some users may prefer the action of an AI model, while others may appreciate the contradictory answers of the other.
Human attachment is a relatively new concept for some AI models that have not been well understood. For example, recently hundreds of people in San Francisco Performed a funeral For the AI model of anthropic, Cloud 3.5 sonnet, when it was taken offline. In other cases, AI chatbots are contributing to mentally unstable people Mental rabbit hole.
It seems that Openai has more work to align its AI model for personal user preferences.
