A precise projection of solar system output is critical for efficient power grid operation and effective energy management. Now, forecasting solar radiance is a crucial part of this projection process. It may be done in a number of ways, such as by using AI-based techniques and remote sensing data with physical models. The recent literature explored the potential various sequential models such as LSTM and its variants for solar radiation forecasting. However, there is not much research done on how well the sequential models perform when forecasting solar radiation across different prediction horizons. In this context, this study performs the performance assessment of some state-of-the-art sequential models such as LSTM, BiLSTM, Hybrid LSTM (proposed model), Simple RNN and Stacked LSTM for solar radiation forecasting over varying forecast horizon (h = 5 h to h = 45 h). The experimental results highlight that although the stacked LSTM, Hybrid LSTM and BiLSTM generally demonstrated better performance for most horizons and performance metrics, the proposed hybrid model has been dominated most of the time.