Gravar-mail: Balancing trust and utility in Large Language Models: a comprehensive trade-off analysis of key performance metrics