OpenAI and Google's AI models both passed Tokyo University's top-tier entrance exam, but showed different strengths—OpenAI scored perfectly on math

OpenAI and Google each developed AI models that achieved top-tier marks on the University of Tokyo's entrance exam for the most selective program (理3, or Science III). OpenAI's model scored a perfect 100 on the math section, while the two AIs showed varying performance across subjects.

The test results expose how AI systems excel at different tasks: despite both passing at elite levels, each model demonstrated distinct strength patterns, suggesting that today's leading AI systems are not equally capable across all domains—particularly mathematics versus other subjects.

For students and job-seekers, this benchmark signals that AI cannot yet reliably replace human judgment on high-stakes exams that require nuanced reasoning across diverse fields. For AI researchers and companies, it highlights the gap between narrow excellence (perfect math scores) and broad, consistent performance needed for real-world problem-solving.