When OpenAI launched GPT-5 last week, the company said the model would simplify the ChatGPT experience by serving as a unified AI system that could automatically select the best way to respond to user prompts. The goal was to eliminate the need for a complex model picker – a menu CEO Sam Altman has called cumbersome.
But GPT-5 has not delivered the seamless experience OpenAI hoped for. On Tuesday, Altman said on X that the company has introduced new “Auto,” “Fast,” and “Thinking” settings for GPT-5, letting users either rely on the model’s router or bypass it entirely.
Paid subscribers can also once again access several legacy models – including GPT-4o, GPT-4.1, and o3 – after their removal last week prompted backlash from users attached to their specific styles and personalities. GPT-4o now appears in the model picker by default, while others can be re-enabled in 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 wrote. He added that OpenAI plans to expand per-user customization of model personalities.
The rollout has been bumpy. GPT-5’s router was reportedly malfunctioning on launch day, leading some to believe the model underperformed compared with previous versions. Altman addressed these concerns in a Reddit AMA, while OpenAI’s VP of ChatGPT, Nick Turley, said the company is iterating quickly.
Routing prompts to the right model is technically complex, requiring split-second decisions to balance speed, quality, and user preferences. Some users value verbosity, others prefer concise or contrarian responses — highlighting that attachment to specific AI models is a new and not well-understood phenomenon.
The reaction mirrors other instances of public sentiment toward AI models, such as a San Francisco gathering held to “mourn” Anthropic’s retired Claude 3.5 Sonnet. For OpenAI, GPT-5’s launch underscores the challenge of aligning AI performance with both functional needs and emotional expectations.
