Multimodal clinical tasks for evaluating how agents observe, reason, use tools, and answer in medical settings.
🌟 Overview
MedCTA is a benchmark for evaluating clinical tool agents. Each example contains a clinical image, a clinical user query, a reference tool-use trajectory, and a final ground-truth answer.
MedCTA evaluates whether agents can:
