This survey provides a comprehensive review of text summarization research, covering statistical methods, deep learning, pre-trained language models (PLMs), and large language models (LLMs). It is organized into two parts: (1) an overview of datasets, evaluation metrics, and summarization methods before the LLM era, including traditional statistical methods, deep learning approaches, and PLM fine-tuning techniques; and (2) a detailed examination of recent advancements in benchmarking, modeling, and evaluating summarization in the LLM era. The survey discusses research trends, open challenges, and proposes promising research directions in summarization, aiming to guide researchers through the evolving landscape of summarization research. Text summarization is a critical task in NLP, involving the distillation of the most important information from a source to produce an abridged version. With the rise of the Internet, summarization research has gained significant attention. Early efforts in automatic text summarization (ATS) date back to the 1950s, with unsupervised feature-based systems emerging in the 1990s and early 2000s. In the 2010s, the focus shifted to training deep learning frameworks in a supervised manner, leveraging large-scale training data. More recently, the advent of self-supervised pre-trained language models (PLMs) such as BERT and T5 has significantly enhanced summarization performance through the 'pre-train, then fine-tune' pipeline. This progression culminates in the current era dominated by large language models (LLMs). LLMs have revolutionized both academic NLP research and industrial products due to their ability to understand, analyze, and generate texts with vast pre-trained knowledge. By leveraging extensive text data, LLMs can capture complex linguistic patterns, semantic relationships, and contextual cues, enabling them to produce high-quality summaries that rival those crafted by humans. However, the rise of LLMs has also triggered an existential crisis among NLP researchers, with some arguing that summarization is (almost) dead. This survey aims to provide a comprehensive overview of the state-of-the-art summarization research in the new era of LLMs, discussing research trends, open challenges, and promising research directions. The survey categorizes existing approaches and discusses problem formulation, evaluation metrics, and commonly used datasets. It systematically analyzes representative text summarization methods prior to the LLM era, encompassing traditional statistical methods, deep learning approaches, and PLM fine-tuning techniques. It also synthesizes insights from recent summarization literature during the LLM era, discussing research trends and open challenges, and proposes promising research directions in summarization. This survey seeks to foster a deeper understanding of the advancements, challenges, and future prospects in leveraging LLMs for text summarization, ultimately contributing to the ongoing development and refinement of NLP research.This survey provides a comprehensive review of text summarization research, covering statistical methods, deep learning, pre-trained language models (PLMs), and large language models (LLMs). It is organized into two parts: (1) an overview of datasets, evaluation metrics, and summarization methods before the LLM era, including traditional statistical methods, deep learning approaches, and PLM fine-tuning techniques; and (2) a detailed examination of recent advancements in benchmarking, modeling, and evaluating summarization in the LLM era. The survey discusses research trends, open challenges, and proposes promising research directions in summarization, aiming to guide researchers through the evolving landscape of summarization research. Text summarization is a critical task in NLP, involving the distillation of the most important information from a source to produce an abridged version. With the rise of the Internet, summarization research has gained significant attention. Early efforts in automatic text summarization (ATS) date back to the 1950s, with unsupervised feature-based systems emerging in the 1990s and early 2000s. In the 2010s, the focus shifted to training deep learning frameworks in a supervised manner, leveraging large-scale training data. More recently, the advent of self-supervised pre-trained language models (PLMs) such as BERT and T5 has significantly enhanced summarization performance through the 'pre-train, then fine-tune' pipeline. This progression culminates in the current era dominated by large language models (LLMs). LLMs have revolutionized both academic NLP research and industrial products due to their ability to understand, analyze, and generate texts with vast pre-trained knowledge. By leveraging extensive text data, LLMs can capture complex linguistic patterns, semantic relationships, and contextual cues, enabling them to produce high-quality summaries that rival those crafted by humans. However, the rise of LLMs has also triggered an existential crisis among NLP researchers, with some arguing that summarization is (almost) dead. This survey aims to provide a comprehensive overview of the state-of-the-art summarization research in the new era of LLMs, discussing research trends, open challenges, and promising research directions. The survey categorizes existing approaches and discusses problem formulation, evaluation metrics, and commonly used datasets. It systematically analyzes representative text summarization methods prior to the LLM era, encompassing traditional statistical methods, deep learning approaches, and PLM fine-tuning techniques. It also synthesizes insights from recent summarization literature during the LLM era, discussing research trends and open challenges, and proposes promising research directions in summarization. This survey seeks to foster a deeper understanding of the advancements, challenges, and future prospects in leveraging LLMs for text summarization, ultimately contributing to the ongoing development and refinement of NLP research.