MedBench v5 Adds Hallucination Tracking and Dynamic Stress Tests for Clinical AI

MedBench v5 evaluates clinical multimodal models across 63 tasks using 14 cognitive sub-dimensions and 4 agent environments, with built-in hallucination propagation monitoring across four interaction stages and three information-flow stressors. It targets silent hallucination failures that static QA benchmarks miss.