Infino claims top VectorDB benchmark speed at 1M vectors, lower cost at 1B

Infino, a vector database engine, has posted the lowest latency results on a standard benchmark at the one-million-vector scale, outperforming comparable engines at most recall levels. The engine is designed to operate across both small, memory-resident workloads and large, object-storage-backed deployments without requiring two separate systems. As a dataset grows, Infino shifts its internal structure from an in-memory graph index to an object-storage index, while keeping the same query API and binary. At billion-vector scale, the company estimates monthly costs of around $2,784 on object storage, compared to roughly $14,000 for a fully RAM-resident setup. The post was written by a member of the Infino team, and the benchmarks were conducted on Infino's own engine.