KYODO NEWS




 

The artificial intelligence ChatGPT earned perfect scores in nine subjects in this year's unified university entrance examinations, marking its first full score, an AI venture said Tuesday.

When LifePrompt Inc. tested the generative AI chatbot for the weekend's university exams, the chatbot's latest version had an accuracy rate of 97 percent across 15 subjects, including the nine.

The nine subjects with full scores included mathematics, chemistry, informatics, and politics and economy. The generative AI chatbot performed worst in Japanese language, with a score of 90 percent, according to the Tokyo-based company.

Students take the annual exams when applying to national and some private universities, choosing some of the subjects. Roughly 500,000 people applied for the two-day tests nationwide, with the results to be used by 813 universities, colleges and junior colleges for their screening processes.

For subjects typically chosen by applicants to the University of Tokyo's highly competitive Human Sciences I program, ChatGPT scored 97 percent, higher than 66 percent in 2024 and 91 percent in 2025. LifePrompt has been testing the generative AI since 2024.

A score of 89 percent provides a 50 percent possibility of being accepted by the Human Science I in this year's exams, according to major cram school Kawaijuku Group.

"Due to improvement in AI's processing and reading abilities, it scored full marks in mathematics. In Japanese language test, it seems it still cannot adequately organize information written in Japanese," said Satoshi Endo, head of LifePrompt.

With access to search engine websites blocked, the AI chatbot answered the questions using only data stored itself.