A recent benchmark from Machine Learning Mastery compares pgvector with dedicated vector databases, focusing on what a developer might expect from a typical laptop. The headline finding is that both approaches report latencies in the low single-digit to tens of milliseconds range. In other words, for many prototyping or small-scale workloads, the choice of vector store may not be the dominant factor in response time.
The benchmark is careful to note that these numbers are not fixed. Exact latency depends heavily on the CPU, available RAM, and whatever else happens to be running on the machine at the time. That means the results are best read as a rough guide rather than a definitive ranking—two laptops with different specs could easily produce different winners.
Because this is a single source, it is worth treating the comparison as indicative rather than conclusive. The article does not claim that pgvector is always as fast as a dedicated vector database, nor does it suggest that dedicated systems are unnecessary. Instead, it highlights that on modest hardware, the performance gap may be small enough that other considerations—such as operational simplicity, existing PostgreSQL infrastructure, or feature requirements—can drive the decision. Production environments with larger datasets, higher concurrency, or specialized indexing needs may tell a different story.{