• OpenAI reinstates GPT-4o after backlash over GPT-5’s underwhelming launch and abrupt removal of older models.
  • CEO Sam Altman admits rollout mistakes, promises warmer updates, new usage modes, and clearer transition periods for future upgrades.

OpenAI’s much-anticipated GPT-5 launch has turned into a lesson in managing user expectations. The Microsoft-backed AI company, which had hyped the model as a major leap forward, is now backtracking on some decisions after a wave of criticism from its user base.

Soon after GPT-5 went live on August 7, OpenAI discontinued older models, including GPT-4o. That move triggered strong backlash, with many ChatGPT Plus subscribers threatening cancellations. 

CEO Sam Altman admitted the company mishandled the transition.

“I think we totally screwed up some things on the rollout,” he told reporters last week.

To calm users, OpenAI reinstated GPT-4o under a new “Legacy models” option for paying customers. Other older versions, such as o3 and GPT-4.1, are also accessible through advanced settings. Altman promised that future model changes will include clearer transition periods.

The GPT-5 debut was supposed to showcase improvements in coding, reasoning, accuracy, and multimodal abilities. Instead, users complained of shorter, less engaging responses, with some calling the model “dumber” than GPT-4o. 

In response, OpenAI has started rolling out behavioral updates to make GPT-5 feel “warmer” and more conversational. “Changes are subtle, but ChatGPT should feel more approachable now,” the company said in a post on X.

Nick Turley, product head of ChatGPT, acknowledged that GPT-5’s blunt style didn’t land well with many.

“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,” he explained.

The new model now offers three usage modes: Auto, Fast, and Thinking, allowing users to choose between speed and depth of responses. Plus and Team subscribers have also received higher usage limits, with up to 3,000 weekly messages in Thinking mode.

Despite criticism, OpenAI says demand remains strong.

“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,” Altman noted, underscoring the scale of the rollout challenge.

With ChatGPT’s weekly user base nearing 700 million, OpenAI’s latest misstep shows just how delicate product changes have become when an AI model is woven into the daily routines of millions.

Edited by Harshajit Sarmah