How a 2021 Quantization Algorithm Quietly Outperforms Its 2026 Successor

The EDEN vector quantization algorithm from 2021 consistently outperforms its 2026 successor, TurboQuant. The primary difference lies in EDEN's use of an analytically derived optimal scale factor, a step that TurboQuant-mse omits, making it a degenerate case of EDEN. This optimal scaling allows EDEN-biased to achieve lower Mean Squared Error across various bit-widths and dimensions. Furthermore, the unbiased variant, EDEN-unbiased, substantially surpasses TurboQuant-prod, often matching its accuracy while using one less bit per coordinate. These performance gains are demonstrated on standard benchmarks for tasks like nearest-neighbor search and apply to use cases such as KV-cache compression and distributed training.