Enhancing Answer Reliability Through Inter-Model Consensus of Large Language Models