Does 80% of 256k perform the same as 80% of 1M+?

A user on reddit's dev community is asking about the performance of Astra's deep retrieval. Specifically, they want to know if 80% utilization of a 256k window (approximately 200k tokens) performs similarly in retrieval and formatting accuracy to 80% utilization of its maximum 1.05M window (approximately 840k tokens). They are seeking findings from anyone who has benchmarked Astra at both scales in a desktop environment.

This query specifically compares the performance of Astra's deep retrieval at 80% utilization across two distinct window sizes, 256k and 1.05M, unlike general performance inquiries.