VectorDBBench
A benchmarking tool for vector databases and cloud vector search services, covering insertion, vector search, filtered search and full-text search under one visual interface.
VectorDBBench runs repeatable benchmark workloads against vector databases and cloud vector services, covering insertion, vector search, filtered search and, more recently, BM25-style full-text search, reporting recall, queries per second and load metrics for each run. Its registered datasets range from classic corpora such as SIFT and GIST to Cohere and OpenAI embeddings and multimodal datasets hosted on Hugging Face, at scales up to one hundred million vectors.
An intuitive visual interface lets users start benchmarks and view comparative result reports without needing to be benchmarking specialists themselves, and it produces cost-effectiveness reports aimed specifically at comparing cloud vector services. Registered Hugging Face datasets can be benchmarked directly through the tool's existing performance test case.
The project is sponsored by Zilliz, the company behind the open-source Milvus vector database, and is released under the MIT license.