When Openai launched GPT-5 last week, the company said the model would simplify the chatgpt experience. Openai hoped that the GPT-5 would act as a kind of “one size fits all” AI model with a router that would automatically decide how to best answer user questions. The company said this overall approach would eliminate the need for users to navigate its model selector – a long, complicated menu with AI settings that Openai CEO Sam Altman has publicly said he hates.
But it seems that the GPT-5 is not the total AI model Openai hoped it would be.
Altman said in a post on X on Tuesday that the company introduced new “Auto”, “Fast” and “Thinking” settings for GPT-5 that all Chatgpt users can choose from Model Picker. The Auto setting seems to act as GPT-5S model router, which Openai originally announced; However, the company also gives users opportunities to bypass it, giving them access to fast and slowly similar AI models directly.
Alongside GPT-5’s new states, Altman said that paid users can again access more older AI-Models-Inclusive GPT-4O, GPT-4.1 and O3-AS was printed just last week. The GPT-4O is now in the model picker by default, while other AI models can be added from Chatgpt’s settings.
“We are working on an update to GPT-5’s personality that should feel warmer than the current personality, but not as annoying (for most users) as GPT-4o,” wrote Altman in the post about X. “But a learning for us from the last few days we are really just coming to a world with more per-user adaptation of model personality.”
Chatgpt’s model picks now appear to be as complicated as ever, suggesting that the GPT-5s model router does not have universally satisfied users as the company hoped. Expectations for GPT-5 were Sky High, with many hoping Openai would push the boundaries of AI models, as it had with the launch of GPT-4. However, GPT-5’s rollout has been harder than expected.
The printing of the GPT-4O and other AI models in Chatgpt triggered a setback among users who had been linked to the AI models’ answers and personalities in ways that Openai had not expected. In the future, Altman says the company will give users plenty of prior notice if it ever secretes GPT-4O.
TechCrunch -event
San Francisco
|
27-29. October 2025
GPT-5S Model Router also seemed to be largely broken on the launch day. It made some users feel that the AI model was not as performing as previous Openai models, and forced Altman to solve the problem in an AMA session on Reddit. However, it seems that the GPT-5 Router may still not be satisfactory for all users.
“We don’t always want to get everything on trial # 1, but I’m very proud of how fast the team can iterate,” Openais VP for Chatgpt, Nick Turley, wrote in a post of X Tuesday.
Routing PROMPS to the right AI model is a difficult task that requires customizing an AI model to a user’s preferences, as well as the specific question they ask. The router then has to make a decision on which AI model should be sent promptly in just a few seconds -that way, if a prompt goes to a quick similar AI model, the answer can still be fast.
More broadly, some people show preferences for AI models that go beyond quick or slow answers. Some users like the verbs of an AI model, while others may appreciate another counter -counter.
Human connection to certain AI models is a relatively new concept that is not well understood. For example, hundreds of people in San Francisco recently held a funeral for Anthropics AI model, Claude 3.5 Sonnet when it was taken offline. In other cases, AI -Chatbots appear to contribute to mentally unstable people walking down psychotic rabbit holes.
It seems that Openai has more work to do with adjusting its AI models to individual user preferences.