Long context in models acts as working memory allowing for processing and recalling large amounts of data accurately. The precision of recalling from long context is crucial, enabling the consideration of massive volumes of data and context like entire books, films, audio, or code bases. A longer context window, such as a million tokens, facilitates reasoning over and retrieving information from the entire corpus of interest, enabling new use cases that are not feasible with shorter contexts. Despite the advantages, the computational expense of processing large context windows poses a challenge, especially in scenarios like analyzing entire movies or textbooks, leading to increased processing power requirements and costs.