Auto text summarization is the challenge of generating brief and eloquent text with no human intervention meanwhile retaining the key elements of the original content. This paper performs thorough analysis of existing extractive and abstractive text summarization algorithms on datasets of BBC and DAWN news articles and proposes a Deep Learning architecture which achieves state of the art accuracy and also helps in considerable reduction of training time. The research uses sequence to sequence based models like LSTMS, Transformers and their variants like BERT, T5 and GPT to analyze the performance of complex architectures on the task of summarizing news articles. Finally, we use the models learned for summarization of BBC news articles to generate the headline for the DAWN news articles. We use ROGUE, BLEU as evaluation measure for extractive summarizers and WEEM4TS and some other word embedding based measures for abstractive summarizers. The research was successful in achieving the performance closers to the bench while requiring relatively lesser training time.