Many companies have been promoting their models' capability to handle long context. For those outside the companies, a context of 1 million tokens still seems somewhat magical or requires enormous ...