Performance of Cloud-Hosted Large Vision-Language Models on the Japanese National Examination for Clinical Laboratory Technicians: Comparative Benchmarking Study

Kota Murakami, Ryosuke Matsuzawa, Yuya Tahara-Arai, Haruka Ozaki, Stefano Brini · 2026

Read Performance of Cloud-Hosted Large Vision-Language Models on the Japanese National Examination for Clinical Laboratory Technicians: Comparative Benchmarking Study by Kota Murakami, Ryosuke Matsuzawa, Yuya Tahara-Arai, Haruka Ozaki, Stefano Brini (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13601776.txt.

Description

Kota Murakami and colleagues put cloud-based AI models to a hard test in this benchmarking study from twenty twenty-six. The test was Japan's national exam for clinical laboratory technicians, which mixes text questions with image questions. The team built a set of eight hundred questions from recent exams and gave them to six large vision-language models, with no extra training.

The best models cleared the exam's sixty percent pass mark by a wide margin, and GPT-5 scored highest. Giving the models the images raised scores for the top systems, though one smaller model gained nothing. The team also checked two hundred fresh questions from an exam held after the models' training cutoff. Scores there held steady, which suggests the models were reasoning, not recalling leaked answers. The study offers a shared yardstick for judging AI on mixed text-and-image medical exams.

Catalog record

CREATORKota Murakami, Ryosuke Matsuzawa, Yuya Tahara-Arai, Haruka Ozaki, Stefano Brini
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE57 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC13601776

Browse the full library · About the Tower of Babel