OpenAI’s latest flagship model, GPT-5, was supposed to simplify ChatGPT by removing the need for users to choose between multiple AI models.
Instead, it has revived the very feature it aimed to kill — the model picker — and added even more complexity.
GPT-5’s big promise falls short
When GPT-5 launched last week, OpenAI touted it as an all-in-one AI system equipped with a built-in “router” that would automatically decide how best to answer user questions.
The Auto mode works like the original router, while Fast and Thinking give users direct access to different performance levels — a move that effectively bypasses the unified approach.
Old favorites return after backlash
In a surprise twist, paid subscribers can once again use several older AI models, including GPT-4o, GPT-4.1, and o3. GPT-4o is now a default option, while the others can be enabled via settings. These models were only deprecated last week, sparking user backlash that caught OpenAI off guard.
Altman acknowledged the criticism, promising that if GPT-4o is ever retired again, users will get plenty of advance notice.
