The language of the research Marathi is a morphologically rich language that does not have dedicated research in Grammatical Error Correction (GEC) although other languages have made significant progress in NLP. The reason of this gap can be mainly explained by the impossibility of publicly available datasets and benchmarks of Marathi GEC. To fill this gap, we have done the extensive review of Marathi NLP literature, comparing datasets and language models as well as language issues. We also examined GEC projects in other languages to determine effective methodology and state-of-the-art models that can be used in the Marathi. Through these observations, we have built a high-quality dataset of incorrect-correct sentence pairs in a systematic way, and we were initially interested in basic Subject-Object-Verb (SOV) types of sentences. To assess how well three transformer-based models, IndicBART, Marathi-T5, and Varta-T5 perform in Marathi GEC, we optimized them on this dataset. The performance analysis based on the BLEU measure shows that IndicBART and Varta-T5 are much higher than Marathi-T5 with a BLEU score of 0.98. Moreover, both models are characterized by the high precision and recall, which means that they are accurate on detecting and correcting grammatical mistakes of the predetermined scope of the dataset. This study will give Marathi GEC a basis that will not only offer a benchmark set, but also a robust one at that. Baseline models, which are a part of further elaboration of Marathi NLP applications.