In a stunning development reflecting the accelerating pace of technological progress, the famous artificial intelligence model “ChatGPT”, developed by “OpenAI”, achieved a historic milestone by scoring higher than the highest scores achieved by top human students in the 2026 entrance exams for the Universities of Tokyo and Kyoto. These two universities are among the oldest, most selective, and most competitive academic institutions in Japan, making this achievement a major turning point in the evolution of artificial intelligence capabilities and its analytical and cognitive levels.
Test methodology and evaluation details
This interesting announcement came last Monday from “LivePrompt”, a company specializing in artificial intelligence projects and research based in the Japanese capital, Tokyo. The tech company explained that it tested the new “ChatGPT 5.2 Thinking” model by providing it with complex exam questions after converting them into image data for the model to process and understand. To ensure absolute accuracy and reliability of the results, essay answers were graded and evaluated by a group of expert teachers and specialists at “Kawaijuku”, one of the largest and most renowned preparatory schools in Japan’s education sector.
Exceptional results at the University of Tokyo
In the University of Tokyo’s “Natural Sciences 3” track exam, which is widely viewed as the most competitive and difficult academic track in the country as it paves the direct path to joining the prestigious medical school, the artificial intelligence proved overwhelming superiority leaving no room for doubt. ChatGPT scored 503 points out of a possible 550, outperforming by a massive margin of about 50 full points the highest score achieved by the best human student taking the same exam, who only managed 453 points. The achievement did not stop at this dazzling feat, as the model also managed to get a perfect score in the complex mathematics section.
As for the “Humanities and Social Sciences” exam, the model continued its academic brilliance, scoring 452 points out of 550, thereby exceeding the highest human score recorded in the “Humanities and Social Sciences 3” track, which at the time reached only 434 points.
Dominating Kyoto University exams
The scene was not much different in the halls of the venerable Kyoto University, where the model demonstrated superior and exceptional capabilities in handling advanced exams. In the rigorous Law Faculty exam, the artificial intelligence garnered 771 points, easily surpassing the highest human passing score, which stood at 734 points. In an even more complex track representing the Medical Faculty exam, the model set a new record by scoring 1176 points, clearly outperforming the best human applicant whose score stopped at 1098 points.
Remaining challenges and weaknesses
Despite the strong and astonishing performance in subjects requiring quantitative, computational, and analytical capabilities, detailed results showed that the technical model still faces some real difficulties in dealing with open-ended essay questions requiring historical narrative and critical analysis. It only managed to score 25 percent in the essays for the world history exam. Conversely, the model proved very high proficiency in foreign languages, achieving 90 percent on the English exam.
A chronology of unprecedented evolution
These striking results cap a series of rapid and successive improvements in AI language model capabilities. In 2024, LivePrompt tested the “GPT-4” model on University of Tokyo entrance exams, and back then the smart model failed to reach the minimum passing score required to pass. In 2025, the newer model called “o1” managed to cross the passing threshold for the first time in its history, marking the beginning of a new era. As for this year’s result, represented by absolute superiority over all smart human applicants, it represents a massive and historic qualitative leap that only a few experts in the field expected.
This academic achievement was preceded by a strong indicator earlier last January, when LivePrompt proved that the “ChatGPT 5.2 Thinking” model had indeed managed to capture perfect scores in 9 out of 15 subjects within Japan’s Common Test for University Admissions, including mathematics, chemistry, informatics, politics, and economics, achieving an astonishing overall accuracy rate of 97 percent.
The future of educational assessment and the business sector
These discoveries and figures have reignited the philosophical and practical debate about the true purpose of standardized exams in schools and universities. In this context, Satoshi Kurihara, a prominent professor at Keio University and president of the Japanese Society for Artificial Intelligence, stated that humans and AI should not be compared on the same competitive playing field. Explaining his idea, he said: “Just as modern calculators can perform complex calculations faster and more accurately than humans can, it is very natural for AI to score high on these tests.” The university professor added that the time has truly come to fundamentally rethink entrance exams, which in their current familiar format focus merely on testing the ability to recall memorized knowledge and simple calculation abilities.
For his part, Satoshi Endo, a senior official at LivePrompt, issued an urgent call to companies and institutions to pay close attention to these rapid technological developments, concluding by saying: “Given the rapid pace of AI technology evolution, companies will inevitably need to adopt these technologies with a clear, forward-looking vision of what business operations and management will look like 10 to 20 years from now.”
Frequently Asked Questions
- What is the model that excelled in the 2026 Japanese university exams?
The “ChatGPT 5.2 Thinking” model developed by OpenAI excelled in the university entrance exams, achieving higher scores than the best human students. - What universities had their exams used to test the model?
The model was tested on the entrance exams of the Universities of Tokyo and Kyoto, which are among the oldest and most difficult universities in Japan. - Did the AI excel in all types of questions?
No, despite its overwhelming superiority in mathematics sections and complex analytical subjects, the model faced noticeable difficulties in open-ended essay questions, such as world history essays where it only scored 25 percent. - How have AI exam results evolved in recent years?
In 2024, the model failed to pass the minimum passing threshold, in 2025 it managed to pass for the first time, while in the 2026 exams it achieved a massive leap and outperformed all advancing human students.
