OpenAI restores GPT-4o default, adjusts GPT-5 access for users
OpenAI has reinstated GPT-4o as the default large language model (LLM) for all its paying ChatGPT subscribers, a decision that reverses a recent shift following the debut of GPT-5. This change affects users across ChatGPT Plus, Pro, Team, Enterprise, and Edu tiers, who will no longer need to navigate a “show legacy models” setting to access GPT-4o. The announcement, made by OpenAI CEO and co-founder Sam Altman, also included a pledge that if GPT-4o is ever removed again, users will receive “plenty of notice.”
This reversal comes on the heels of a challenging first week for GPT-5, which rolled out on August 7 in four variants: regular, mini, nano, and pro. The new flagship model, intended to supersede GPT-4o, was met with a mixed reception and notable infrastructure issues. Early adopters reported a dysfunctional “autoswitcher” that misrouted prompts, inconsistent performance when compared to the outgoing GPT-4o, and widespread frustration over the sudden deprecation of familiar older models.
With the latest update, paying ChatGPT subscribers will now find GPT-4o automatically selected as their primary model. Furthermore, a new “Show additional models” setting will be enabled by default, restoring direct access to GPT-4.1, along with the reasoning-focused LLMs o3 and o4-mini. All these model options are conveniently located within the “picker” menu, accessible at the top of the ChatGPT session screen across web, mobile, and other applications.
Beyond the reinstatement of GPT-4o, Altman’s announcement detailed new controls for GPT-5 itself. Users can now choose between “Auto,” “Fast,” and “Thinking” modes for the latest model. The “Thinking” mode, designed for more extensive and reasoning-intensive tasks, features a substantial 196,000-token context window. However, this powerful mode is subject to a cap of 3,000 messages per week for paying subscribers. Once this limit is reached, users can continue their work using the lighter “GPT-5 Thinking mini” mode. Altman indicated that these usage limits are subject to change, depending on evolving usage trends. It is worth noting that GPT-4.5, another advanced model, remains exclusively available to Pro users, primarily due to its significant GPU processing demands.
Looking ahead, Altman also hinted at further refinements for GPT-5, including a planned “personality tweak.” The aim is to make the model feel “warmer” than its current default, while simultaneously avoiding the polarizing tone that some users associated with GPT-4o. In the long term, OpenAI is exploring the implementation of per-user customization options. This strategic move could address the deep emotional attachments and strong preferences many users have developed for specific AI models, a phenomenon that has become increasingly apparent with the rapid evolution of these technologies.
These immediate changes are poised to placate a segment of the user base that felt alienated by the abrupt transition to GPT-5 and the removal of their preferred, older LLMs. However, the rapid succession of model changes and the company’s responsiveness to user sentiment also underscore the complex, almost personal, relationship many users are forming with these advanced AI systems, potentially fostering even stronger emotional fixations on specific iterations.