Reasoning models like DeepSeek-R1 improve large language models by training them to produce structured thought processes before answering, making outputs more transparent and easier to audit—though true reasoning and understanding remain beyond their capabilities.