Couchbase, Inc., the developer data platform for critical applications in the AI era, has announced new benchmark results demonstrating the exceptional performance of its Hyperscale Vector Index (HVI). Using the industry-standard VectorDBBench methodology, Couchbase achieved over 700 queries per second (QPS) with sub-second latency and higher accuracy — performing up to 350 times faster than MongoDB Atlas under identical conditions.
The independent test compared Couchbase and MongoDB across datasets of 100 million and 1 billion vectors, measuring QPS, latency, and recall accuracy. Couchbase consistently delivered superior performance, particularly at higher recall levels crucial for AI-driven applications.
At optimized speed settings, Couchbase achieved 19,057 QPS with 28-millisecond latency and 66% recall accuracy, versus MongoDB’s 6 QPS and 62.6-second latency, a 3,000x performance edge. When configured for high accuracy, Couchbase maintained 703 QPS at 93% recall, while MongoDB dropped to 2 QPS at 89% recall with over 40 seconds of latency — a 350x advantage.
“Performance determines whether AI applications deliver real value,” said BJ Schaknowski, CEO of Couchbase. “At billion-vector scale, architectural choices matter. Couchbase gives enterprises the speed, accuracy, and scalability they need without trade-offs — all at a lower total cost of ownership.”
Couchbase’s architecture, leveraging DiskANN and Vamana algorithms with scalar quantization (SQ4), enables high-performance vector search at scale. Testing was conducted on AWS using equivalent hardware configurations for both platforms.
Couchbase 8.0, featuring HVI capabilities, is now generally available for self-managed and Capella deployments across on-premises, cloud, and edge environments.
To access the full benchmark report, visit www.couchbase.com/products/vector-search.
