A new query engine in pgrust v0.4-beta compiles analytical SQL into machine code, and its authors say it ran up to three times faster than ClickHouse on ClickBench. Gravity also supports the full Postgres SQL dialect, making this an ambitious new entrant in database performance, not just another benchmark chart looking for an audience.
Watch Desk analysis
What happened
Gravity is a multi-stage query compiler for the pgrust database project. Its authors describe using a Lisp-based intermediate representation and just-in-time compilation to turn SQL queries into machine code. They say the engine reaches 80 to 90 per cent of the performance of their handwritten Rust reference code.
The implementation includes function inlining, late materialisation of dictionary-encoded strings, and register-based run collapse. The authors also say Claude helped generate highly optimised handwritten Rust baselines during development. Their performance figures are reported by the project’s authors, not an independent comparison.
Why it matters
A database that can run familiar Postgres SQL while compiling analytical queries directly to machine code is an intriguing proposition. If Gravity’s reported results hold up on other workloads, it could offer users another route to fast analytical queries without requiring them to rewrite everything around a new query language.
The comparison with ClickHouse is attention-grabbing, but “up to three times faster” is a best-case figure from a named benchmark, not a general promise. Database performance depends on the query, data and hardware. The useful next question is how the engine behaves beyond the benchmark and how much of its performance advantage survives real applications.
Our read
The interesting story is the combination of Postgres compatibility and a compiler built to squeeze more out of analytical SQL. The benchmark gives Gravity a reason to be noticed; it does not yet settle where it belongs in a production stack. We’d be curious, and keep the database migration plan in the drawer for now.
What to watch
- Whether the authors publish more results across varied workloads and hardware.
- How Gravity performs in practical pgrust use beyond ClickBench.
- Whether the project’s compiler approach and Postgres compatibility continue to develop.
Discussion spark: What would you need to see beyond a ClickBench result before trusting a new query engine with a real workload?
Sources and evidence
- Programming at Light Speed: Building the Gravity Query Engine (29 October 2026, 18:25 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.