StyleBench: Evaluating thinking styles in Large Language Models