OpenAI listens to users: GPT-4o returns and new GPT-5 modes let you control speed, depth, and personality of responses.
OpenAI is rolling out significant updates to ChatGPT, giving users more say in how they interact with its newest GPT-5 model while also restoring popular older models like GPT-4o. The changes come after a wave of user backlash over the company’s recent decision to make GPT-5 the default without warning, replacing GPT-4o—a model many had grown attached to.
In a post on X, OpenAI CEO Sam Altman announced that paid subscribers can now choose between three modes for GPT-5: Auto, Fast, and Thinking. The Auto setting will likely be the default for most users, automatically selecting the best balance between speed and detail. Fast mode prioritizes quick replies, while Thinking mode takes more time to deliver deeper, more detailed reasoning.
For those opting for GPT-5 Thinking, there is now a weekly rate limit of 3,000 messages, after which users will still have access via GPT-5 Thinking mini. Altman noted that these limits could change over time based on how the feature is used. The model now supports a 196k-token context limit, allowing it to handle much longer conversations and documents—an upgrade especially useful for extensive coding projects or large-scale content creation.
One of the biggest updates is the return of GPT-4o, now available again to paid users through the “Show additional models” toggle in ChatGPT’s web settings. This option also brings back GPT-4.1, o3, and GPT-5 Thinking mini, giving users more flexibility in selecting the model that best suits their needs. GPT-4.5 remains exclusive to Pro subscribers due to its higher GPU requirements.
Altman also revealed that OpenAI is working on refining GPT-5’s personality. Many users had complained that GPT-5’s responses felt “colder” and less conversational compared to GPT-4o. While GPT-4o’s more casual tone was appreciated by many, it also drew criticism for being overly agreeable. The goal with GPT-5, Altman said, is to make it warmer and more engaging without going to the extremes of its predecessor.
When GPT-5 was first launched, OpenAI introduced multiple versions—GPT-5, GPT-5 Thinking, and GPT-5 Thinking Pro—and initially planned to let the system decide which to use based on the query. This approach also meant older models like GPT-4o were suddenly removed, sparking frustration from long-time users who preferred the personality and interaction style of those models. Some even threatened to cancel their subscriptions if GPT-4o wasn’t restored.
The new changes mean users can now manually select both the model and the reasoning mode. For GPT-5, the modes offer clear use cases: Auto for balanced performance, Fast for speed, and Thinking for tasks requiring complex reasoning and detailed explanations. This level of control, paired with the return of legacy models, is OpenAI’s attempt to rebuild trust and offer a more user-friendly experience.
Altman’s message was clear—OpenAI is listening. By bringing back GPT-4o and introducing a more transparent way to choose between models and modes, the company hopes to address the concerns of its community while continuing to develop GPT-5 into a more capable and relatable AI assistant.