When Openai launched the GPT-5 last week, the company said the models would simplify the ChatGPT experience. Openai wanted the GPT-5 to function as a kind of “fitting all sizes” AI model, using a router that automatically determines how to best answer user questions. The company said this unified approach eliminates the need for users to navigate model pickers. This is a long, complicated menu of AI options that Openai CEO Sam Altman said he disliked.
However, the GPT-5 appears to have been wanted by the integrated AI model Openai.
In a post on Tuesday in X, Altman said the company has introduced new “automatic”, “fast” and “thinking” settings for the GPT-5. The autoconfiguration appears to work like the GPT-5 model router originally announced by Openai. However, the company also offers users the option to avoid it, allowing them to access responding AI models quickly and slowly directly.
In addition to the new mode of GPT-5, Altman said that paid users will be able to access several legacy AI models again, including GPT-4O, GPT-4.1 and O3, which were deprecated last week. The GPT-4O comes in the model picker by default, but other AI models can be added through the ChatGPT settings.
“We’re working on updating the GPT-5 personality, which should feel warmer than the current personality, but it’s not as annoying as the GPT-4o (for most users) in X’s post.
ChatGpt’s model pickers seem to be as complicated as ever, suggesting that the GPT-5 model routers do not universally meet users as the company wanted them to. Expectations for the GPT-5 are high, and many people were hoping that with the release of the GPT-4 it would push the limits of AI models. However, the GPT-5 rollout was rougher than expected.
The blame for the GPT-4O and other AI models in CHATGPT caused repulsion among users attached to the responses and personality of the AI model in ways Openai didn’t expect. In the future, Altman says that if he condemns the GPT-4o, the company will provide users with sufficient advance notice.
TechCrunch Events
San Francisco
|
October 27th-29th, 2025
The model GPT-5 router also appeared to be largely broken on the launch day. This forced Altman to address the issue in Reddit’s AMA session, feeling that the AI model was less performant than the previous OpenAI models. However, the GPT-5 routers still seem to be unsatisfied for all users.
“We don’t always get everything in Try #1, but we’re extremely proud of how quickly our team can iterate,” wrote ChatGpt’s Vice President of Openai in a post on Tuesday.
Routing the prompts to the right AI model is the difficult tasks that require the AI model to be tailored to the user’s preferences and the specific questions they are asking. The router then needs to decide that the AI model will send a prompt in just a second. That way, if the prompts go to a faster AI model, the response may be faster.
More broadly, some people show preferences for AI models beyond fast or slow responses. Some users like the redundancy of one AI model, while others evaluate the opposite answer to another AI model.
Human attachment to specific AI models is a relatively new concept that is not well understood. For example, hundreds of people in San Francisco recently filmed offline funerals for humanity’s AI model, Claude 3 Sonnet. Otherwise, AI chatbots appear to contribute to people suffering from mental health issues that go down the hole in unhealthy rabbits.
Openai seems to have a lot of work on matching AI models to individual users’ preferences.
Source link
