The paper introduces AraBERT, a transformer-based model specifically trained for Arabic language understanding. The authors address the challenges of morphological richness and limited resources in Arabic NLP tasks such as Sentiment Analysis (SA), Named Entity Recognition (NER), and Question Answering (QA). By pre-training BERT on a large-scale Arabic corpus, AraBERT achieves state-of-the-art performance on most tested Arabic NLP tasks. The model is evaluated on three downstream tasks: SA, NER, and QA, using datasets that cover Modern Standard Arabic (MSA) and dialectal Arabic (DA). The pre-training setup includes Masked Language Modeling (MLM) and Next Sentence Prediction (NSP) tasks, with a pre-trained dataset of 70 million sentences. The paper also discusses the impact of sub-word unit segmentation and fine-tuning strategies for each task. The results show that AraBERT outperforms multilingual BERT and previous state-of-the-art models, highlighting the effectiveness of monolingual pre-training for Arabic language understanding. The models are publicly available on GitHub to encourage further research and applications in Arabic NLP.The paper introduces AraBERT, a transformer-based model specifically trained for Arabic language understanding. The authors address the challenges of morphological richness and limited resources in Arabic NLP tasks such as Sentiment Analysis (SA), Named Entity Recognition (NER), and Question Answering (QA). By pre-training BERT on a large-scale Arabic corpus, AraBERT achieves state-of-the-art performance on most tested Arabic NLP tasks. The model is evaluated on three downstream tasks: SA, NER, and QA, using datasets that cover Modern Standard Arabic (MSA) and dialectal Arabic (DA). The pre-training setup includes Masked Language Modeling (MLM) and Next Sentence Prediction (NSP) tasks, with a pre-trained dataset of 70 million sentences. The paper also discusses the impact of sub-word unit segmentation and fine-tuning strategies for each task. The results show that AraBERT outperforms multilingual BERT and previous state-of-the-art models, highlighting the effectiveness of monolingual pre-training for Arabic language understanding. The models are publicly available on GitHub to encourage further research and applications in Arabic NLP.