Green Vectors vs Elastic BBQ: A Head-to-Head Benchmark

Morphos AI benchmarked Green Vectors against Elastic's Better Binary Quantization (BBQ), one of the most widely used vector compression methods, using the complete Project Gutenberg dataset. The benchmark measured three configurations: Green Vectors alone, BBQ alone, and Green Vectors combined with BBQ.

Relevancy score: closer to 1 is more accurate.

The challenge

A next-generation vectorization technology must be tested against established industry solutions. Elastic BBQ is a powerful compression method for large-scale vector workloads. The benchmark measured not only storage but the often-overlooked metrics of accuracy and latency, at the scale of the full Project Gutenberg corpus.

The results

Green Vectors alone achieved 1.5GB of storage at a .9658 relevancy score. BBQ alone required 175GB at a .4576 relevancy score. That makes Green Vectors roughly 116 times more storage-efficient and more than twice as accurate as BBQ on this benchmark.

Layering, in context

Green Vectors and quantization operate on different axes, so they can be layered. In this benchmark, however, Green Vectors alone was the optimal configuration. Combining Green Vectors with BBQ held relevancy at .9653 but used 2.6GB, more storage than Green Vectors alone at 1.5GB, because on an already-minimal index BBQ's rotation and correction overhead costs more than it saves. The practical takeaway is that Green Vectors delivers the efficiency quantization aims for, which makes a separate quantization step optional.

Why this matters

The key finding is that Green Vectors achieves its efficiency while improving accuracy, where compression alone sacrifices accuracy to save space. Green Vectors eliminates redundancy at the source rather than degrading precision.