ClickHouse Launches PostgresBench, an Open Benchmark for Postgres-Compatible Services

ClickHouse published PostgresBench on 20 June 2026, an open-source benchmarking project hosted on GitHub that covers more than 40 databases and provides fully reproducible methodology — public queries, datasets, and results included.

The project extends ClickHouse's existing benchmarking work. ClickBench, the company's OLAP benchmark, already tracks 40-plus databases using transparent, reproducible methodology. PostgresBench applies a similar philosophy to the Postgres ecosystem specifically, targeting the crowded field of Postgres-compatible managed services and extensions where performance claims are common but apples-to-apples comparisons are hard to come by.

Reproducibility is the operative word here. Publishing the queries, datasets, and raw results means any operator can rerun the suite against their own infrastructure and verify or challenge the numbers. That matters in a benchmark landscape where vendor-commissioned tests and cherry-picked workloads have made many practitioners rightly skeptical. The GitHub repository makes the methodology auditable at the source level, not just at the summary level.

This is not ClickHouse's first time benchmarking adjacent territory. In February 2026, the company published findings showing that pg_clickhouse — a Postgres extension that delegates analytics queries to a ClickHouse backend — posted the fastest analytics results on ClickBench among tested Postgres extensions. In January 2025, ClickHouse ran a broad JSON workload benchmark pitting ClickHouse against MongoDB, Elasticsearch, DuckDB, and PostgreSQL across a billion-document dataset. PostgresBench fits into that pattern: a series of open benchmarks, each scoped to a specific competitive surface.

In May 2026, ClickHouse also released CostBench, an open benchmark focused on cost-performance trade-offs across cloud data warehouses. Taken together, the benchmarking portfolio now spans OLAP throughput (ClickBench), JSON and semi-structured data (the billion-docs suite), cloud warehouse economics (CostBench), and now the Postgres service layer (PostgresBench). Each benchmark is, at minimum, a positioning tool. Each also provides genuine reference data for engineers making infrastructure decisions.

Worth flagging: ClickHouse has a direct commercial interest in several of the comparisons these benchmarks surface. pg_clickhouse, for instance, routes Postgres queries to ClickHouse — so a benchmark that ranks it highly is also product marketing. That does not invalidate the methodology, but engineers using PostgresBench results to inform procurement decisions should replicate the runs on representative workloads of their own, not treat published numbers as a substitute for internal testing. Reproducibility being the explicit design goal here at least makes that verification tractable.

The Postgres service market has grown substantially as organizations have leaned into managed database offerings from AWS, Google Cloud, Azure, Neon, Supabase, PlanetScale, and others. Many of these services carry the Postgres protocol while differing significantly in storage engines, replication architecture, and query execution. A benchmark that covers 40-plus of them with consistent methodology gives DBAs and platform engineers a baseline reference that has been absent from this space. Whether PostgresBench achieves that status will depend on community uptake and how diligently ClickHouse keeps the dataset and query set current as the service landscape evolves.

The repository is publicly available on GitHub at ClickHouse/PostgresBench.