Recent assessments reveal that artificial intelligence models from OpenAI, Google, and Anthropic have outperformed human candidates on the challenging entrance exam for the University of Tokyo, particularly in scientific disciplines.
According to reports from Japanese media, AI systems have shown remarkable advancements in academic performance. The models developed by OpenAI and Google achieved scores exceeding those of the top human applicants in the most difficult sections of the entrance examination.
This evaluation was conducted by an AI startup in collaboration with an educational organization, which tested contemporary models, including GPT, Gemini, and Claude, against actual exam questions. Educators assessed the responses based on real examination tasks.
In the scientific section, OpenAI scored 503 out of a possible 550 points, while Google followed closely with 496 points. In contrast, the highest score achieved by a human candidate was 453 points. Some AI models also attained perfect scores in mathematics and specific natural science subjects.
The strengths of these AI systems were particularly evident in the natural sciences and engineering. However, their performance in the humanities was inconsistent; the models excelled in physics and chemistry but faced challenges in history and the Japanese language.
Researchers noted a significant improvement in the quality of AI responses compared to the previous year. Experts emphasize that each model exhibits unique strengths influenced by their training data and architectural design.
AI models from OpenAI and Google have demonstrated superior performance on the University of Tokyo's entrance exam, surpassing top human scores, particularly in scientific subjects. This trend highlights the rapid advancements in AI capabilities across various academic disciplines.