OpenAI’s latest artificial intelligence model, GPT-5, was introduced recently with promises to make ChatGPT easier to use and more capable. The company envisioned the new model as a single, all-purpose AI that could decide on its own the best way to respond to any question. But the rollout has not gone exactly as planned.

The “model router” approach was meant to replace the long list of selectable AI models, a menu that CEO Sam Altman has said he dislikes.

Older OpenAI models make a return

On Tuesday, Altman announced on X that GPT-5 now comes with a trio of settings, “Auto,” “Fast,” and “Thinking”, all available in the selection menu. The Auto mode works like the original router system, choosing responses automatically. However, users can now bypass it and pick between faster or slower, more thoughtful AI responses directly.

Altman also revealed that several retired AI models are making a return for premium users. GPT-4o, GPT-4.1, and o3, all removed from ChatGPT just last week, are now available again. GPT-4o is visible by default in the menu, while others can be enabled through settings.

“We are working on an update to GPT-5’s personality which should feel warmer than the current personality but not as annoying (to most users) as GPT-4o,” Altman posted. He added that feedback from recent days showed a need for more personalized model personalities for each user.

The changes mean ChatGPT’s model selection menu is still lengthy, despite earlier plans to simplify it. GPT-5 was expected to build on the success of GPT-4, but its debut has faced challenges.

One early controversy came when the company removed older AI models, sparking backlash from users who preferred their unique response styles. Some had grown attached to the tone and mannerisms of those models. Altman said future removals, such as GPT-4o, will be announced well ahead of time.

ChatGPT launch day glitches add to frustration

Another issue came on launch day, when GPT-5’s router appeared to malfunction. As Cryptopolitan previously noted, users thought that the model didn’t perform as well as earlier versions. Altman addressed the concerns in a Reddit AMA, but some users remain unsatisfied with the new system.

Nick Turley, OpenAI’s vice president of ChatGPT, acknowledged the bumpy beginning in a post on X: “We’re not always going to get everything on try no.1 but I am very proud of how quickly the team can iterate.”

Routing user prompts to the right AI model is a complex challenge. The system must consider both a person’s preferences and the nature of their question, then choose an AI model within a fraction of a second.

But user preferences often go beyond response speed. Some people enjoy more detailed, wordy answers, while others appreciate models that take contrarian stances.

In fact, user connections to AI models have shown unexpected depth. Recently, hundreds of people in San Francisco held a mock funeral for Anthropic’s Claude 3.5 Sonnet model after it was shut down. In other cases, chatbots have been linked to unhealthy online behavior, with vulnerable individuals getting drawn into troubling thought patterns.

For now, GPT-5’s mission to be the universal AI solution is a work in progress. OpenAI’s challenge will be to balance a streamlined interface with the growing demand for personal choice — and to match user expectations and needs.

If you're reading this, you’re already ahead. Stay there with our newsletter.