My friends benchmarked managed vector databases under production-like conditions: high-throughput ingest, concurrent queries, filtering, and read–write mixed workloads.

The post includes the methodology, the dataset, and the open-source tool they published for running the benchmarks.