From Rejected to Top of the Class in Just Two Years: ChatGPT Surpasses Human Students at Japan’s Elite Universities | WPN World People News

Japanese AI startup LifePrompt recently released test results showing that after using OpenAI’s latest generative model to take the entrance examinations for The University of Tokyo and Kyoto University, the AI not only achieved scores high enough for admission but, in some departments, even surpassed the highest scores of human applicants. The results have sparked discussions about AI capabilities and the future of education systems.

The experiment used OpenAI’s latest model, “ChatGPT-5.2 Thinking.” The research team converted exam questions into image format and had the AI answer them. Essay questions were graded by instructors from the cram school organization Kawaijuku Educational Institution according to official scoring standards, and the results were then combined with the standardized test scores.

For the The University of Tokyo exams, out of a total score of 550 points, the AI scored 452 points in the humanities track and 503 points in the science track, both exceeding the university’s published highest admitted scores. In the highly competitive Science Category III track, the AI scored about 50 points higher than the actual top-ranked student, drawing particular attention.

At Kyoto University, the AI scored 771 points on law-related examinations, surpassing the highest human score of 734 points. In medical-related programs, it achieved 1,176 points, again exceeding the top human applicant’s score.

Among all subjects, the improvement in mathematics attracted the most attention. According to the analysis, the AI had still lost significant points on the difficult mathematics section of the University of Tokyo science exam last year, but it is now capable of achieving perfect scores on highly challenging problems. Its English scores also reached a very high level. However, the AI still struggled with subjects such as world history that require extended analysis, contextual reasoning, and structured argumentation, resulting in relatively lower scores.

The research team noted that large language models have made significant advances in calculation, knowledge organization, and reasoning abilities. However, when it comes to constructing viewpoints, analyzing historical contexts, and organizing essays, they still lag behind humans.

Notably, LifePrompt has previously challenged entrance exams from Japan’s top universities using AI several times. In 2024, the version of ChatGPT used at the time still failed to meet the admission threshold. It was only after the model was updated the following year that it first passed the standard. Now, in just two years, AI has progressed from failing admission to achieving top-ranked scores, demonstrating that the development speed of generative AI has far exceeded public expectations.

In response to these results, academics in Japan have begun discussing the future direction of education. Some experts believe that as AI increasingly demonstrates advantages in knowledge-based and problem-solving examinations, traditional admission systems may need to reconsider their evaluation criteria and place greater emphasis on areas where humans still hold advantages, such as creativity, critical thinking, and the ability to define problems.