Qdrant open-sources a production-scale vector search benchmark: billions of vectors, thousands of RPS

qdrant_engine · x · 2026-09-01

Qdrant argues public benchmarks don't come close to production conditions — billions of vectors, thousands of RPS, tight tail latency — so it built and open-sourced its own: corpus and hosting from Hugging Face, compute from Vultr, distribution via SkyPilot, plus the Supernova code and dataset.

They say the era of 1M-vector datasets and synthetic approximations is over; evaluation should use real data with real ground truth and open tooling.

Related event: Qdrant Open-Sources Supernova Benchmark for Billion-Scale Vector Search(2 posts)→

Original post →

More from Infra

Infra channel →