Accuracy of a Vision-Language Model on Challenging Medical Cases