Tokyo: Japanese media outlet Nikkei reported on Tuesday that an artificial intelligence (AI) model performed remarkably well on Japan's 2026 national university entrance examination. OpenAI's GPT-5.2 Thinking model secured first place with perfect scores in nine subjects.
According to Thai News Agency, the test was jointly developed by Nikkei and LifePrompt, a Japanese AI startup. It used the latest models from major AI companies like OpenAI and Google to test 15 core subjects in the university entrance exam held on January 17-18. OpenAI's latest model, ChatGPT 5.2 Thinking, achieved an average score of 96.9 out of 100 in 15 subjects and a perfect score in 9 subjects. In comparison, Google's Gemini 3.0 Pro scored 91.4.
The Japanese university entrance examination covers 21 subjects across 7 fields of study. The average score for test-takers in the 15 most popular subjects in 2026 is expected to be around 58.1 points.
Previously, earlier OpenAI models had already taken such tests, with average scores increasing from 66 in 2024 to 91 in 2025, reflecting the rapid advancements in artificial intelligence performance.
The analysis also indicates that the latest artificial intelligence models performed well in mathematics, physics, chemistry, and biology, but scored lower in Japanese language and geography.
Although artificial intelligence can correctly interpret geometric shapes in mathematics exams, it still makes mistakes on questions involving world maps, reflecting the limitations of AI in recognizing and interpreting irregular and complex visual data.