If an artificial intelligence model such as ChatGPT had taken the entrance exams for Japan¡¯s top university in 2026, it would have been assessed as top of the class and admitted for scoring higher than any human test takers, a study by AI startup LifePrompt has found.

The research used three major AI models ¡ª ChatGPT 5.2 Thinking by OpenAI, Gemini 3 Pro Preview by Google and Claude Opus 4.5 by Anthropic ¡ª and had them take the actual entrance exam used by the University of Tokyo in February 2026 to assess candidates for courses set to start in April.

The university¡¯s category 3 science exam, often taken by those who want to enter the institution¡¯s medical school, is considered the most difficult exam to pass in Japan.

Two of the AI models surpassed the highest score recorded for a human test taker in this year¡¯s exam, which was around 454 out of the total 550 points. ChatGPT, with the highest score, got around 504 points, beating out the human by over 50 points.

These exams are notable in that they primarily require long-form written responses rather than a multiple-choice answer, which has previously posed a burden for AI models. The 2026 scores marked major progress for the computerized test-takers, given that just two years ago, in 2024, ChatGPT-4 could not reach passing grades in any category at the University of Tokyo.

In fact, this year ChatGPT scored full marks on all the institution¡¯s mathematics exams, including the one for science majors. Just last year, ChatGPT o1 was only able to score 38 out of the 120-point total.

In its latest attempt, ChatGPT scored near full marks in subjects in science fields such as physics and chemistry, and scores remained very high in English as well, at around 90%. However, it struggled more with tests in humanities fields including long-form answers for world history, where the bot was only able to score around 25%.

The study found that the AI models continue to face difficulties interpreting visual diagrams embedded in the tests, and are unable to construct a coherent flow in written Japanese.

Nonetheless, ChatGPT scored around 503 out of the total 550 points for all three science courses at the University of Tokyo, and around 453 out of the same total for the three humanities courses ¡ª all well above the passing grade and higher than any human test taker. Gemini performed slightly better than ChatGPT in the humanities field, scoring around 460 out of the 550-point total.

The same study found that the three AI models could also pass the entrance exam for another of Japan¡¯s top universities, Kyoto University, with scores higher than the maximum score obtained by a human test taker across all majors.