The fast growth of data has transformed text processing, making it challenging to extract key information efficiently. Text summarization techniques address this by reducing lengthy documents while retaining essential content. Automatic text summarization can be broadly categorized into two main types, extractive and abstractive summarization. In extractive summarization, the final summaries are constructed by selecting and extracting content directly from the source text. On the other hand, abstractive summarization takes a different approach. It aims to understand the source text and convey its core ideas in a more concise form using linguistic techniques. Arabic is spoken by over 300 million people and serves as the official language in 22 countries. There is a growing demand for effective Arabic summarization systems to facilitate efficient information processing and retrieval in the Arab-speaking world. Transformers revolutionize NLP by using self-attention to capture long-range dependencies and process input sequences at the same time, improving efficiency. In abstractive summarization,Transformers play an important role because they produce clear, logical summaries that go beyond simply selecting important passages and rewriting the text in a way that is human-like. In this paper, we present a comprehensive investigation of Arabic summarization datasets and techniques introduced to date, with a focus on fine-tuning and using pre-trained transformer models for Arabic summarization, such as AraT5 and AraBERT. We compare their performance using the ROUGE metric on the Wikilingua multi-sentence dataset and find that AraT5 outperforms AraBERT, showing its effectiveness in abstractive summarization tasks.