After ChatGPT-5 went online yesterday, many users discovered that OpenAI had deactivated the GPT-4o version, causing huge controversy. A reporter from the "Kechuangban Daily" noticed that there were a lot of complaints about the poor usability of GPT-5 on social media, and some users even called out "Give me back GPT-4o".
In the face of user controversy, OpenAI announced today that it has re-launched GPT-4o for Plus and Team users. If you need to use it on multiple platforms, users can enable "Show old model" in the settings of the ChatGPT web version to access GPT-4o. At the same time, mini versions of GPT-5 and GPT-5 thinking will be launched next week.
OpenAI CEO Sam Altman also posted two tweets on the X platform in response to the controversy caused by the offline GPT-4o and the online launch of GPT-5.
Altman admitted frankly that he had underestimated users' love for GPT-4o, and said that this has strengthened the company's concept of providing customized services for different users, and will launch more customized content in the future than currently. At the same time, it announced that it will allow Plus users to continue to use GPT-4o and will closely monitor user usage and consider the retention period of older versions of the model.
In response to the negative feedback from users about GPT-5, Ultraman explained that this was due to the damage to the automatic switcher yesterday, which caused GPT-5 to perform abnormally. He said that in the future, he will focus on completing the deployment and improvement of GPT-5 and ensuring its stability.
Altman also revealed that after GPT-5 went online, ChatGPT API traffic doubled within 24 hours, and OpenAI is making every effort to optimize the system and find more capacity. But he admitted that he still expected to face serious capacity challenges next week.
A reporter from "Kechuangban Daily" observed that users' "complaints" about GPT-5 mainly focused on the fact that although the new model improved the professionalism of answers, it lacked the warmth, personality and imagination of emotional interaction.
Some users reported to a reporter from the Science and Technology Innovation Board Daily that "GPT-5 is less humane and will not take the initiative to care or comfort me." In contrast, "GPT-4o can infer whether I am scared, nervous, or uneasy through my words and tone, and provide corresponding comfort."
According to Wang Xiangjie, chief computer analyst at the Southwest Securities Research Institute, the emotional differences reported by users may be due to factors such as model versions and optimization word adjustments, and can be corrected at any time. He pointed out that at this stage, GPT-5 does have insufficient emotional expression in the text writing section, and the optimization focus is on enhancing the professionalism of the model. This may also be the strategic focus adopted by OpenAI in this version.
Wang Xiangjie believes that from the development history of OpenAI, its product ideas still focus more on memory and personality as the core. "The next version will definitely be integrated back." "Personality has always been the core goal pursued by OpenAI, which is to make AI a companion to the user, rather than a professional answer."