ChatGPT has achieved perfect scores in nine subjects in Japan’s unified university entrance examinations for the first time, underlining the rapid gains in the capabilities of generative artificial intelligence, Kyodo News reports.
Tests conducted by Tokyo-based AI venture LifePrompt found that the latest version of the chatbot achieved an overall accuracy rate of 97% across 15 subjects in the examinations held over the weekend. Full marks were recorded in nine areas, including mathematics, chemistry, informatics, and politics and economics.
According to Kyodo, performance was weakest in the Japanese language paper, where the system scored 90%, suggesting continued limitations in organising and synthesising complex written material in Japanese. By contrast, advances in processing speed and reading comprehension drove a notable improvement in quantitative subjects, particularly mathematics.
The annual two-day examinations are a key gateway to higher education in Japan, used by 813 universities, colleges and junior colleges. Around 500,000 candidates sit the tests each year when applying to national universities and some private institutions, selecting subjects relevant to their chosen courses.
For subjects typically taken by applicants to the University of Tokyo’s highly competitive Human Sciences I programme, ChatGPT recorded a score of 97%. This compares with 66% in 2024 and 91% in 2025, highlighting a sharp year-on-year improvement since LifePrompt began testing the system in 2024. By comparison, an overall score of 89% is estimated by the Kawaijuku Group, a major cram school operator, to give applicants a 50% chance of admission to the programme this year.
During the tests, access to search engines and external websites was blocked, meaning the chatbot relied solely on its internal data and reasoning capabilities.
No comments:
Post a Comment