GPT-5 rollout blends modes with safety concerns and legacy pushback

OpenAI’s GPT5 was unveiled as a unifying, more capable AI intended to power ChatGPT for hundreds of millions of users. In the days since its release, the rollout has been punctuated by a combination of notable improvements, user pushback over removing legacy models, and a wave of scrutiny over safety and reliability. The company moved quickly to address concerns by restoring access to GPT4o for paid subscribers, tweaking model options, and outlining ongoing efforts to tune how GPT5 interacts with users, all while signaling that this is a rapidly evolving product.

A central feature change has been the introduction of distinct GPT5 modes designed to balance speed, depth, and reasoning. Auto lets the model decide how long to “think” based on the prompt, Fast prioritizes quick responses, and Thinking gives higher reasoning power at the cost of slower replies. In a further adjustment, the Thinking mode comes with a weekly limit of 3,000 messages, after which users can switch to a lighter Thinking mini tier. OpenAI has also expanded user control by adding a model picker that separates GPT5 into Auto, Fast, Thinking, and legacy options such as GPT4o, GPT4.1, o3, and o4mini for those...