Spelling error correction in Arabic language using transformer models represents a recent challenge in Natural Language Processing (NLP). This study tackles this issue by fine-tuning the AraT5v2 – base –1024, an Arabic Text-To-Text Transformer model. We present our findings to demonstrate the effectiveness of our approach. We employed the Arabic Grammar Corrections dataset for training our model. Our evaluation strategy included testing the model on various subsets of this data, specifically 5K, 20K, Test200, and a final AraSpell dataset. On the 5K and 20K datasets, the model achieved a character error rate (CER) of 0.39% and a word error rate (WER) of 0.42%. Compared to previous studies, the fine-tuned AraT5v2 – base –1024 model demonstrated superior accuracy on both the Test200 and AraSpell datasets, achieving a CER of 0.50% and a WER of 0.51% on Test200, and a CER of 0.61% and a WER of 0.81% on AraSpell, thereby outperforming other models in spelling correction.