Exact Match (EM) requires the model's answer to be character-for-character or structurally identical to the reference answer to be counted as correct. It measures correctness in the strictest sense and is commonly used in tasks such as question answering, code generation, and structured prediction. It is typically reported as a percentage or a score between 0 and 1, where higher values indicate better performance. It is not lenient toward partially correct answers or synonymous rephrasings, and is often used in conjunction with Token F1 or semantic metrics.