Managing the enormous volume of textual data that is currently available requires text summarization. Because of the increasing demand for effective summary techniques across several languages, this work focuses on abstractive text summarization in Assamese. The objective is to evaluate how well two popular recurrent neural network (RNN) architectures—Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM)—perform in producing precise and cohesive summaries. This was accomplished by creating a dataset of 15,000 Assamese articles and the summaries that went with them. The data was cleaned using preprocessing methods such stopword removal, tokenization, and stemming. Then, using this dataset, LSTM and GRU models were created and trained. ROUGE scores, which evaluate the quality of the generated summaries by contrasting them with reference summaries, were used in the evaluation. The results show that the LSTM model produces better Assamese summaries than the GRU model. The improved performance of LSTM can be ascribed to its capacity to store important information across lengthy sequences and capture long-term dependencies.