PlanetScale Put Full-Text Search Inside Postgres With TIN, Claiming 541x GIN Throughput and Killing the Elasticsearch Sync Path

PlanetScale (primary; vendor-run benchmarks, corroborated by runtimewire and byteiota)·medium signalXBlueskyLinkedInCopy link

PlanetScale's TIN ("Text INdex") reached GA v1.0.2 for all Postgres and Neki databases on September 16 and hit the HN front page at 218 points on September 19. Published benchmarks: 25x throughput and 26x lower p99 versus ParadeDB on mixed queries, 36x versus pg_textsearch on disjunctions with concurrent writes, and 541x throughput with 1,356x lower latency versus Postgres GIN on conjunction and phrase queries, plus 10,260 QPS against ParadeDB's 291 on in-memory disjunction COUNT. The category consequence is the second cluster: an application can search a just-committed row without waiting for a sync into Elasticsearch or OpenSearch, which removes the reason most teams pay for search as a separate service.