Normative Reasoning in Large Language Models: A Comparative Benchmark from Logical and Modal Perspectives