Sparse vectors have tens of thousands of dimensions and...

Dense vectors require a fixed 768 coordinates per chunk, leading to high storage costs (e.g., 33 GB for ten million chunks). Sparse vectors, despite having high-dimensional ambient spaces, only store non-zero entries (e.g., 160 per chunk), resulting in significantly lower storage (e.g., 2 GB for the same volume). The sparse cost scales with document length, but the dense cost is constant and generally much higher.