Couchbase has announced benchmark results showing its Hyperscale Vector Index (HVI) significantly outperformed MongoDB Atlas in billion-scale AI workloads, boasting up to 350 times faster performance and 93 percent recall accuracy.
The independent test, conducted using the industry-standard VectorDBBench tool, evaluated both databases across datasets containing up to one billion vectors.
Couchbase reportedly achieved over 700 queries per second (QPS) with sub-second latency, while MongoDB delivered just two QPS at comparable accuracy levels.
Couchbase said its performance edge comes from its use of DiskANN and Vamana algorithms, enabling efficient nearest-neighbor searches across distributed memory and disk partitions.
At optimized speed settings, the company recorded 19,057 QPS with 28-millisecond latency and 66 percent recall, compared to MongoDB’s six QPS and 62.6-second latency.
CEO BJ Schaknowski said the findings highlight how database architecture directly impacts the scalability and responsiveness of enterprise AI applications.
The tests were performed on identical Amazon Web Services infrastructure, and Couchbase 8.0—with the new HVI capabilities—is now generally available for on-premises, cloud, and edge deployments.
Leave a Reply