GoKawiilParadeDB doubles COUNT query speed over TantivyIndex in new benchmark
ParadeDB published results showing its Postgres-based search extension now runs COUNT queries roughly twice as fast as TantivyIndex (TIN) on a Wikipedia dataset benchmark, reaching 5,353 queries per second versus 2,645.8 QPS. The company says the gains come from optimizations such as segment partitioning that reduce costly visibility-map and CTID page accesses, which previously consumed over 80% of page reads during COUNT operations.
GoKawiil's interpretation of the reporting above, not reported fact.
The post suggests that for filtered or sorted search queries beyond plain COUNT, ParadeDB claims speedups of up to 66 times, which could matter for teams evaluating Postgres-native search engines against alternatives. Because the comparison is self-published by ParadeDB using its own benchmark harness, the magnitude of real-world gains may vary depending on workload and dataset.
- ParadeDB reports 2x faster COUNT query throughput than TantivyIndex in its benchmark (5,353 vs 2,645.8 QPS).
- The speedup stems partly from segment partitioning that cuts down expensive visibility-map and CTID lookups.
- ParadeDB claims up to 66x faster performance on filtered and sorted search queries compared to simple unfiltered BM25/COUNT cases.
Source: paradedb.com, 2026-10-28
Published there as: “Solving Postgres' Count Problem”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.