Weeks after the bumpy debut of its latest flagship AI model, OpenAI has said it is tweaking GPT-5 to make its AI-generated responses seem warmer and more familiar.
The post-launch changes are based on user feedback that GPT-5’s responses “felt too formal.” “Changes are subtle, but ChatGPT should feel more approachable now. You’ll notice small, genuine touches like “Good question” or “Great start,” not flattery. Internal tests show no rise in sycophancy compared to the previous GPT-5 personality,” OpenAI said in a post on X on Friday, August 15.
The behavioural changes to GPT-5 are expected to roll out in the coming week. It is one of many updates that have been announced by the Microsoft-backed AI startup since the model was launched on August 7. Users have been left disappointed by the underwhelming release of GPT-5, which has been hyped up since the company’s 2023 release of GPT-4.
GPT-5 also suffered several delays due to safety testing and compute limitations. When the model became freely available in ChatGPT this month, users pointed out that the advancements they had been expecting seemed incremental, with GPT-5’s main improvements related to cost and speed. In response to these issues, OpenAI CEO Sam Altman reportedly told journalists at a dinner last week, “I think we totally screwed up some things on the rollout.”
“On the other hand, our API traffic doubled in 48 hours and is growing. We’re out of GPUs. ChatGPT has been hitting a new high of users every day. A lot of users really do love the model switcher. I think we’ve learned a lesson about what it means to upgrade a product for hundreds of millions of people in one day,” Altman was quoted as saying by The Verge.
ChatGPT has quadrupled its user base in a year, and is nearing over 700 million people each week, as per reports.
On GPT-5’s behavioural issues, Nick Turley, the product head of ChatGPT, said, “GPT-5 was just very to the point. I like that. I use the robot personality — I’m German, you know, whatever. But many people do not, and they really like the fact that ChatGPT would actually check in with you.”
Story continues below this ad
Here’s a brief list of all the changes and improvements announced by OpenAI since the launch of GPT-5.
New GPT-5 modes
GPT-5 includes Auto, Fast, and Thinking modes. Fast mode gives users faster answers from GPT-5 while Thinking mode means the model takes more time to give deeper answers. Auto mode routes between Fast and Thinking modes. The three modes can be selected by users within the model picker in ChatGPT.
“GPT-5 will seem smarter starting today. Yesterday, the autoswitcher broke and was out of commission for a chunk of the day, and the result was GPT-5 seemed way dumber,” Altman said in a post on X.
GPT-4o is back
During a Reddit Ask Me Anything (AMA) session, multiple users requested OpenAI to bring back GPT-4o. Replying to the Reddit thread, Altman said that the OpenAI team had heard user feedback and had decided that they will offer an option for Plus users to continue using GPT-4o and “will watch usage to determine how long to support it.”
Story continues below this ad
GPT-4o is available under “Legacy models” by default for paid users. Other legacy AI models such as o3 and GPT-4.1, as well as GPT-5 Thinking mini, can be added to the model picker within ChatGPT by enabling “Show additional models” in ChatGPT’s settings. To be sure, this option is only available for paid users.
Moving forward, Altman said that the company will give users a more clear “transition period” when deprecating AI models in the future, as per a report by TechCrunch.
Increasing usage limits
ChatGPT Plus and Team subscribers now get up to 3,000 messages in a week when using GPT-5 in Thinking mode, with extra capacity on GPT-5 Thinking mini when they hit this limit. The company has also made GPT-5 available for ChatGPT Enterprise and Edu subscribers.
Additionally, OpenAI has said it is working on improvements at the user interface (UI) level so that users can more easily enable Thinking Mode in GPT-5 and more clearly see which AI model is responding to their query or prompt.