<p>In this study, we compare the performance of classical time series models (AR, MA, ARMA, ARIMA) and modern deep learning techniques (LSTM, Bi-LSTM, Seq2Seq) alongside a state-of-the-art Transformers model. These models have been individually studied and applied to specific problems, but the lack of a standardized benchmark for comparing their performance across diverse datasets creates a significant gap in the literature. This research aims to benchmark the performance of representative time-series models across different datasets. We evaluate the models using three publicly available datasets and compare their prediction accuracy using metrics i.e. MSE, RMSE, and MAE. The results show that modern deep learning techniques outperform classical models, with the sim- plest architecture, LSTM, achieving the lowest MSE across all datasets (e.g., for the temperature dataset, LSTM has the lowest RMSE at 0.01, compared to the highest RMSE at 0.21 by ARIMA). Although the Transformers model demon- strated superior performance to classical models, it did not outperform modern techniques. This suggests that for the datasets used in this study, complex Trans- formers architectures may not be necessary to achieve optimal results. This study provides some insights into time series studies.</p>