OpenAI’s latest AI model, GPT-5, praised by CEO Sam Altman as a significant advancement in artificial intelligence, has faced a mixed reception since its release. Altman characterized GPT-5 as a leap forward, stating that “GPT-3 feels like you’re talking to a high school student. GPT-4 feels like a college student. GPT-5 feels like you’ve got a PhD in your pocket for the first time.”
Despite these claims, many users have reported that GPT-5 exhibits instability and confusion in its responses. For example, when asked the number of “b”s in “blueberry,” GPT-5 initially provided an incorrect answer of “three,” which it later corrected only after prompting. Users have expressed their disappointment, particularly regarding the model’s performance on basic math and spelling tasks.
Feedback from experts has also been critical. Noah Giansiracusa, an associate professor of mathematics at Bentley University, acknowledged some improvements but felt that GPT-5’s reasoning and answering capabilities fell short of expectations.
During a recent AMA (Ask Me Anything) session on Reddit, many users voiced frustration over the inability to select their preferred AI model. In response, Altman promised upgrades to enhance system transparency and functionality. When asked about the perceived decline in GPT-5’s writing quality compared to its predecessor, many users agreed with the sentiment.
Addressing the criticism, Altman noted that an issue with the auto-switch feature had temporarily affected performance. He reassured users that GPT-5 would improve from that point forward, stating that it would soon exhibit greater intelligence.
While there are dissenting opinions, some users have praised GPT-5’s capabilities. Ethan Mollick, a professor at the University of Pennsylvania, highlighted the model’s ability to simplify programming tasks and its impressive research capabilities, although it sometimes produces mechanical responses.
Internationally, researchers in China have also weighed in on GPT-5. Although not officially available in the country, they have assessed its performance. Associate Professor Zhang Linfeng from Shanghai Jiaotong University remarked that GPT-5 does not significantly outperform Chinese AI models and lacks revolutionary breakthroughs. However, he acknowledged improvements in reducing the “hallucination” phenomenon and enhancing programming abilities.
In light of the mixed feedback about GPT-5, Altman admitted that the transition from GPT-4 to GPT-5 was not as smooth as hoped, leading to experiences of lower intelligence with the new model. Consequently, OpenAI plans to reintroduce the GPT-4o model to address these concerns.
