TOKYO, Jan. 20 (Xinhua) -- Artificial intelligence (AI) models achieved remarkable results after being used to take on Japan's 2026 Common Test for University Admissions, with OpenAI's model taking top spot with perfect scores in nine subjects, the Nikkei reported on Tuesday.
The experiment was jointly conducted by the Nikkei newspaper and Japanese AI startup LifePrompt. The latest models from major AI companies, including OpenAI and Google, tackled the questions in 15 major subjects from Japan's university entrance examinations held from Jan. 17 to 18.
The latest model GPT-5.2 Thinking developed by U.S.-based OpenAI recorded an overall score rate of 96.9 out of 100 across all 15 subjects, including perfect scores in nine subjects, while Google's Gemini 3.0 Pro scored 91.4, the report said.
Japan's university entrance exam features 21 subjects across seven disciplines. The average score for the 15 most popular subjects among test-takers this year is estimated at 58.1, it added.
OpenAI's earlier models have previously attempted the exams. Their average score rose from 66 in 2024 to 91 in 2025, reflecting rapid advancement in performance.
The analysis showed that the latest AI models performed particularly well in mathematics, physics, chemistry and biology, but lost more points in Japanese language and geography. For example, while the models were able to correctly interpret geometric figures in mathematics, they made mistakes on questions involving world maps, indicating that AI still has limitations in recognizing irregular and complex graphical information. ■
