What Is PostgreSQL 18? The Answer, Explained
By Michael
PostgreSQL 18 is the major release of the PostgreSQL open-source database that the PostgreSQL Global Development Group shipped on 25 September 2025. Its headline change is a new asynchronous I/O subsystem, which the project measured at up to 3× faster reads from storage in some workloads. It also makes major-version upgrades less painful, adds uuidv7(), virtual generated columns and OAuth 2.0 sign-in, and turns on data checksums for new clusters by default. PostgreSQL 18 gets bug and security fixes until 14 November 2030.
What is PostgreSQL 18?
PostgreSQL ships one major version a year, usually in September, and numbers it with a single integer: 15, 16, 17, 18. A major version can change the on-disk format, so moving from 17 to 18 needs an upgrade step (pg_upgrade, or a dump and restore). Between major versions come minor releases such as 18.1 and 18.2. They contain only bug and security fixes, keep the same data format, and install with a restart.
PostgreSQL 18 is the newest major version on postgresql.org's versioning page as of 16 November 2026, which lists 18.7 as its current minor release. Every major version is supported for five years, so 18 is a reasonable default for a new database today.
What's new in PostgreSQL 18
The summary below follows the project's release announcement. The release notes have the full list.
Asynchronous I/O
Earlier versions waited on each read and relied on the operating system's readahead to guess what comes next. PostgreSQL 18 can issue several reads at once and keep working while they complete. Sequential scans, bitmap heap scans and vacuum use it. The new io_method setting picks how: worker (background I/O processes, on every platform), io_uring (Linux only) or sync (the old behaviour).
Faster, less disruptive major upgrades
- Planner statistics survive the upgrade. Before 18, a freshly upgraded cluster had no statistics, and queries could run badly until
ANALYZEhad finished on every table. 18 keeps them. pg_upgradeis faster on databases with many tables and sequences, runs its checks in parallel with--jobs, and gains a--swapmode that swaps data directories instead of copying or linking files.
Queries that use indexes more often
- Skip scan on multicolumn B-tree indexes: an index on
(tenant_id, created_at)can now help a query that filters only oncreated_at, as long astenant_idhas few distinct values. ORconditions in aWHEREclause can be turned into an index lookup.- Faster hash joins, merge joins that can use incremental sort, and parallel builds for GIN indexes (B-tree and BRIN already had them).
For application developers
uuidv7()generates timestamp-ordered UUIDs. New rows land at the end of the index instead of at random places, which is kinder to B-tree indexes and caches than random v4 UUIDs.uuidv4()is added as an alias forgen_random_uuid().- Virtual generated columns are computed when read instead of stored, and are now the default kind of generated column. Stored generated columns can now be logically replicated.
RETURNINGcan return bothOLDandNEWvalues fromINSERT,UPDATE,DELETEandMERGE.- Temporal constraints:
PRIMARY KEYandUNIQUEover ranges withWITHOUT OVERLAPS, and foreign keys withPERIOD. For example, “one booking per room at any moment” can now be a key. - A
casefold()function and aPG_UNICODE_FASTcollation for case-insensitive comparisons, andLIKEon nondeterministic collations.
Authentication and security
- OAuth 2.0 authentication (
oauthinpg_hba.conf), via a validator extension, so database logins can go through single sign-on. - MD5 passwords are deprecated and will be removed in a future release. Use SCRAM.
- A new
ssl_tls13_cipherssetting, FIPS-mode validation, and SCRAM passthrough forpostgres_fdwanddblink.
Operations and observability
EXPLAIN ANALYZEnow shows buffer usage without asking, and counts index lookups per index scan.- Vacuum freezes more pages during normal runs, which means fewer expensive aggressive vacuums later.
pg_stat_all_tablesrecords time spent vacuuming, and I/O and WAL statistics are available per connection. - Logical replication write conflicts are logged and counted, new subscriptions apply with parallel streaming by default, and idle replication slots can be dropped automatically.
- Version 3.2 of the wire protocol, the first new protocol version since PostgreSQL 7.4 in 2003. libpq still speaks 3.0 by default until drivers and poolers catch up.
What to check before upgrading to PostgreSQL 18
- Data checksums are on by default in clusters created by PostgreSQL 18's
initdb.pg_upgradeneeds both clusters to match, so an old cluster without checksums must be upgraded into a new cluster created with--no-data-checksums. - Full-text search and
pg_trgmindexes may need a reindex afterpg_upgrade. Full-text search now uses the cluster's default collation provider rather than always libc. - MD5 password logins still work but are deprecated. Move roles to SCRAM before a release removes MD5.
How long is PostgreSQL 18 supported?
The project supports each major version for five years after its first release. Dates from postgresql.org/support/versioning:
| Version | First release | Final release (end of support) |
|---|---|---|
| PostgreSQL 18 | 25 September 2025 | 14 November 2030 |
| PostgreSQL 17 | 26 September 2024 | 8 November 2029 |
| PostgreSQL 16 | 14 September 2023 | 9 November 2028 |
| PostgreSQL 15 | 13 October 2022 | 11 November 2027 |
| PostgreSQL 14 | 30 September 2021 | 12 November 2026 |
PostgreSQL 14 got its last release on 12 November 2026. A database still on 14 will not get any more security fixes, so plan its upgrade now.
Where you can run PostgreSQL 18 as a managed database
Managed providers add new major versions at their own pace, so check before you pick one. Two examples from Europe:
- HostStack managed Postgres offers PostgreSQL 15, 16, 17 and 18, and 18 is the default for new databases. An existing database can be upgraded to 18 in place from its page or with
hoststack db upgrade-version db_… --to 18(dump and restore, so expect a short interruption). The database docs list the limits. - Scaleway's Managed Database for PostgreSQL lists 17 as its newest version. PostgreSQL 18 is available on its Serverless SQL Database. The details are in our note on Scaleway Postgres 18.
For a wider comparison of managed Postgres providers in Europe, see the Managed PostgreSQL Europe buyer's guide.
FAQ
What is PostgreSQL 18?
PostgreSQL 18 is the major version of the PostgreSQL open-source database released on 25 September 2025. Its main changes are an asynchronous I/O subsystem (up to 3× faster reads from storage in the project's benchmarks), planner statistics that survive major upgrades, skip scan on multicolumn B-tree indexes, uuidv7(), virtual generated columns, temporal constraints and OAuth 2.0 authentication.
When was PostgreSQL 18 released?
PostgreSQL 18.0 was released on 25 September 2025. As of 16 November 2026 the current minor release is 18.7.
How long will PostgreSQL 18 be supported?
Until 14 November 2030. The PostgreSQL project supports each major version for five years, and 18's final release is scheduled for that date.
Can I upgrade from PostgreSQL 17 to 18 with pg_upgrade?
Yes. pg_upgrade supports it, and in 18 it keeps planner statistics, runs its checks in parallel with --jobs and adds a --swap mode. New 18 clusters have data checksums on by default, so if your old cluster has none, create the new one with --no-data-checksums. Full-text search and pg_trgm indexes may need a reindex afterwards.
Is MD5 password authentication removed in PostgreSQL 18?
No, it is deprecated, not removed. MD5 logins still work in PostgreSQL 18 but will be removed in a future release, so move roles to SCRAM authentication.