Reasoning refers to a model's ability to follow logical steps, draw conclusions, apply rules, or solve problems that require more than pattern recall — still debated whether current models truly reason or simulate reasoning. Chain-of-thought prompting, scratchpads, and extended thinking have improved reasoning performance on math and logic tasks. Benchmarks like GSM8K and ARC test different reasoning facets. The distinction between genuine reasoning and sophisticated pattern interpolation remains one of the most active debates in AI research.