20 Sept 2026
An open-weights release with a date, a name and an empty repository
A 600-billion-parameter flagship announced today says its weights arrive on 15 October. The repository named for them currently holds one…
Full-text search that lives inside Postgres rather than in a separate search service — an inverted index, BM25 ranking and a query language, with transactions, replication and backups intact.
TIN — short for Text INdex — is a full-text search extension for Postgres announced by PlanetScale on 16 September 2026 and generally available immediately on its Postgres and Neki databases. The argument for it is architectural rather than about speed: most teams that outgrow Postgres's built-in text search move their search workload into a separate service and then spend years keeping two systems consistent. TIN adds an inverted index type designed for search, BM25 relevance ranking and a query language called TINQL — boolean operators, quoted phrases, fuzzy matching, wildcard prefixes and proximity queries — while behaving like an ordinary Postgres index: it works with complicated WHERE clauses, with replication and with backups, and it maintains correct transaction visibility, so a row that has not committed is not findable. PlanetScale publishes a detailed comparison, and it is the company measuring its own product, which is the first thing to know about the numbers. Against ParadeDB 0.25.2, pg_textsearch 1.4.0 and PostgreSQL 18.6's built-in GIN index, on an 85 GB Stack Exchange corpus of 150 million documents running on a single AWS i7i.8xlarge instance, it reports handling 25 times as many mixed ranked queries per second as ParadeDB, 10 times as many conjunction and phrase queries as ParadeDB and 541 times as many as GIN, and 36 times as many disjunction queries as pg_textsearch while writes were running concurrently, with 10,260 queries a second on a Wikipedia corpus once the index fit in memory. The headline claim is a range — three to thirty times faster than competing Postgres search indexes — and the spread in that range is worth more attention than the peak. No price is published for TIN itself; it is part of the database product. PlanetScale also released Lead, a separate open-source extension under the AGPL that accepts TIN-compatible SQL and returns correct results without a production search index, so the same application queries can run in development, test and CI, offline. Not to be confused with anything else called Tin; this is the Postgres extension.
An earned signal from verification, reviews, awards, transparency and engagement — the vendor can't buy it.
Updated 9/20/2026
No questions yet. Be the first to ask about TIN.
Everything here links back to the same verified catalogue. Pick your next stop.