GPT-5.2 Model Achieves Near-Perfect Score on 2026 Japanese University Entrance Exam

In January 2026, the performance of advanced artificial intelligence models against the rigorous Common Test for University Admissions in Japan established a new benchmark for academic assessment. Reporting on Tuesday, January 20, 2026, the Nikkei newspaper detailed a joint experiment with the Japanese AI firm LifePrompt, which demonstrated the accelerated capabilities of artificial intelligence in high-stakes intellectual performance. OpenAI's GPT-5.2 Thinking model achieved an overall score of 96.9 out of 100 across 15 major subjects examined between January 17 and 18, 2026.

This result contrasts sharply with the estimated human average score for the 2026 cohort in those 15 popular subjects, which was recorded at 58.1. The 2026 performance marks a significant upward trajectory for AI in this specific academic domain, following OpenAI's previous average scores of 66 in 2024 and 91 in 2025. The comprehensive Japanese university entrance structure comprises 21 distinct tests across seven main disciplines, although the experiment focused on 15 key subjects. Google's Gemini 3.0 Pro model also participated, securing a strong overall score of 91.4 in the same assessment.

Analysis of the scores revealed distinct patterns in the AI models' proficiencies, highlighting both areas of strength and current limitations. The models demonstrated near-mastery in quantitative disciplines, including mathematics, physics, and chemistry, achieving perfect scores in nine of the 15 tested subjects. Conversely, measurable weaknesses emerged in areas requiring nuanced interpretation of complex visual data, specifically when interpreting world maps, despite high performance on geometric figures within mathematics. This disparity suggests that while Large Language Models excel at structured reasoning common in STEM fields, challenges persist in visual pattern recognition that deviates from standard numerical or textual formats.

This event continues a historical context of using these entrance exams to benchmark artificial intelligence progress in Japan. The National Institute of Informatics (NII) previously led the To-Robo-kun project, which aimed to assess AI's intellectual labor capacity using the University of Tokyo entrance exam in 2015. Dr. Noriko Arai, the project director at NII, noted that the emergence of To-Robo-kun prompted a shift in educational policy toward free-response questions to better clarify AI's capabilities and deficits. The NII remains a central institution for integrated informatics research advancing Japan's technological frontier.

The implications of GPT-5.2 Thinking's near-perfect score extend beyond academic metrics, signaling a potential shift in global educational paradigms and intellectual labor. The model's ability to surpass the human average by a significant margin establishes a new standard for artificial general intelligence applications, evidenced by its state-of-the-art performance on benchmarks like GPQA Diamond and FrontierMath. The ongoing challenge involves responsibly integrating these powerful systems while addressing the identified gaps in complex spatial reasoning.