The artificial intelligence chatbot ChatGPT has achieved full marks in nine subjects in Japan’s unified university entrance examinations for 2026. This marks the first time the AI system has secured perfect scores in multiple subjects in the highly competitive national-level tests. The performance was revealed after an independent evaluation conducted shortly after the completion of the two-day examinations.
AI Accuracy Reaches 97 Percent Across 15 Subjects
According to testing results released by LifePrompt Inc., the latest version of ChatGPT recorded an overall accuracy rate of 97 percent across 15 subjects. Among these, nine subjects saw full scores, including mathematics, chemistry, informatics, and politics and economy. The chatbot’s weakest performance was observed in the Japanese language paper, where it achieved an accuracy rate of 90 percent, indicating remaining challenges in advanced language structuring and interpretation.
Performance Compared to Previous Years
LifePrompt has been assessing ChatGPT’s performance in Japan’s entrance examinations since 2024. The results indicate a sharp improvement over time. For subjects commonly selected by applicants to the University of Tokyo Human Sciences I programme, ChatGPT scored 97 percent in 2026. This marks a significant rise from 66 percent in 2024 and 91 percent in 2025, reflecting steady advancements in the AI’s reasoning and problem-solving abilities.
Context of Japan’s Unified University Entrance Exams
The unified university entrance examination is a critical component of Japan’s higher education admission process. Each year, approximately 500,000 candidates appear for the exams, selecting subject combinations based on their intended courses and institutions. The scores are used by over 800 universities, colleges, and junior colleges nationwide as part of their screening and admission procedures.
Expert Observations on AI Capabilities
Commenting on the results, representatives from LifePrompt attributed the improved performance to advances in AI processing speed and comprehension abilities. Mathematics, in particular, showed notable gains due to enhanced logical reasoning. However, the Japanese language section highlighted areas where AI still struggles, especially in organising and synthesising complex written information in native linguistic contexts.
Exam Conditions and Acceptance Benchmarks
During testing, ChatGPT was evaluated under strict conditions, with no access to external websites or search engines. All responses were generated solely from data stored within the model. According to benchmarks shared by Kawaijuku Group, a score of 89 percent in the relevant subject set typically corresponds to a 50 percent chance of admission to the University of Tokyo’s Human Sciences I programme. ChatGPT’s performance comfortably exceeded this threshold.
Significance for Education and AI Development
The results highlight the growing capabilities of artificial intelligence in academic assessment environments. While the findings do not suggest the replacement of human examinees, they underline the pace at which AI systems are approaching and, in some areas, surpassing average student performance in standardised testing. Experts note that such developments raise important questions about the future role of AI in education, assessment integrity, and learning support systems worldwide.