GPT-5 was supposed to make ChatGPT simpler. OpenAI’s plan was to use a unified model router that would automatically choose the best AI response for each prompt. Instead, the model picker has returned, now offering modes like “Auto,” “Fast,” and “Thinking,” along with older models such as GPT-4o – available again for paying users.
This added complexity came in response to user backlash over the removal of familiar models and early issues with GPT-5, including a broken router. OpenAI acknowledges the challenge of matching prompts with individual preferences and plans to improve personalization moving forward. Many users form strong attachments to their favorite models, making this task even harder.