This chapter focuses on evaluating large language models (LLMs) and their capabilities across various tasks. It begins by introducing widely used evaluation metrics, encompassing tasks such as language modeling, classification, conditional text generation, question answering, task execution, and more. Different evaluation paradigms are then discussed, such as benchmark-based evaluation, human evaluation, and model-based evaluation. The chapter categorizes LLM abilities into basic and advanced levels. Basic abilities include language generation, knowledge utilization, and complex reasoning. Advanced abilities cover human alignment, environmental interaction, and tool usage. Various datasets and benchmarks, such as MMLU, GSM8K, and HELM, are used to test these capabilities. Challenges like hallucination, outdated knowledge, reasoning inconsistencies, and numerical inaccuracies are highlighted, along with mitigation strategies such as retrieval-augmented generation, chain-of-thought prompting, and model fine-tuning. The chapter concludes by emphasizing the need for continuous improvements in evaluation methodologies to ensure LLMs align with real-world applications and human expectations.