MED-ITA

We introduce MED-ITA, a large-scale evaluation benchmark designed to assess the medical knowledge and understanding capabilities of Large Language Models (LLMs) in the Italian language. The benchmark comprises 10,000 multiple-choice questions drawn from official national examinations used to certify and recruit medical professionals across the Italian healthcare system.
MED-ITA covers 50 distinct clinical and biomedical categories, ranging from core medical sciences (e.g., pathological anatomy, pharmacology and clinical toxicology, clinical pathology) to specialised clinical disciplines (e.g., neurology, cardiac surgery, psychiatry) and cross-disciplinary fields (e.g., legal medicine, health statistics and biometry). All questions are written in authentic Italian and reflect the structure, terminology, and cultural specificity of real-world medical assessments.
MED-ITA offers the first fine-grained, domain-specific benchmark for Italian medical language understanding, providing a foundation for advancing culturally and clinically grounded AI applications in healthcare.