Large Language Model (LLM) evaluation is the systematic process of assessing model performance across multiple dimensions, from factual accuracy and reasoning capabilities to safety, bias, and production efficiency. Evaluation encompasses both offline benchmarks (standardized tests measuring capabilities on fixed datasets) and online methods (human feedback, A/B tests, and real-world performance monitoring). The challenge lies in the multifaceted nature of language understanding: a single metric cannot capture whether a model is truly useful, trustworthy, and production-ready. Effective evaluation requires combining automated metrics with human judgment, as purely computational approaches often miss nuances like factual hallucination, harmful biases, or contextual appropriateness that only humans can reliably detect.