DuckDB 2.0: what's coming, and DuckDB inside dbt

DuckDB has become the go-to tool for analysing files on a laptop: it runs SQL directly on CSV, Parquet and JSON, needs no server, and is remarkably fast. Two things happened recently that make it even more relevant. The DuckDB team previewed DuckDB 2.0 in August 2026, with the release planned for 21 October 2026 (a tentative date), and in September DuckDB started shipping inside dbt v2. The current stable version is 1.5.6. Here is what is coming.

Why analysts like DuckDB

-- Query a folder of Parquet files directly, no loading step
SELECT region, SUM(revenue) AS revenue
FROM 'sales/*.parquet'
GROUP BY region
ORDER BY revenue DESC;

No database to set up, no import, and results in seconds on millions of rows. That is why DuckDB now powers many data tools, notebooks and even parts of cloud platforms.

DuckDB 2.0 highlights

DuckDB as a server

The biggest shift: DuckDB can now share its databases over the network, so a team can query one DuckDB database instead of passing files around. A new CONNECT statement also lets DuckDB send queries to other databases such as PostgreSQL and MySQL.

CALL quack_serve(token = 'my_token');      -- share this DuckDB over the network
CONNECT 'postgres://localhost/mydb';         -- route queries to PostgreSQL

DuckDB stays an embedded, in-process database at heart; server mode is an option, not a replacement.

VARIANT for semi-structured data

A new VARIANT type stores semi-structured data such as JSON from APIs in an efficient, column-like way, so queries that pull out a few fields from large JSON documents get much faster. New json_set, json_insert, json_replace and json_remove functions make editing JSON in SQL easier.

SELECT json_set('{"a": 1}', '$.b', '2');

More SQL features

• Triggers, before and after changes, a first for DuckDB.
• Variables with $name syntax, handy in scripts: SELECT * FROM orders WHERE amount > $threshold;
• Data changes inside CTEs, and NEAREST joins for similarity search.
• A new, more extensible SQL parser that aims to stay compatible with existing queries.

Faster, especially on cloud storage

Asynchronous I/O now runs across the engine for Parquet, CSV and DuckDB's own files, with the biggest gains when reading from network or cloud storage. Smarter aggregation and pruning speed up many queries, and in one recursive query benchmark from the DuckDB team, runtime fell from 4.9 seconds to 0.12 seconds. Built-in time zones and collations replace the ICU extension, with 2.2 to 2.6 times speed-ups on the team's benchmarks.

Breaking changes to expect

DuckDB 2.0 uses a new default storage format, and finishes the move to the new lambda syntax. Files written by 2.0 may not open in older versions, so upgrade everyone who shares database files at the same time, and keep a backup. The team notes that some details may still change before release.

Want to try it early? An alpha is available, but the DuckDB team says it is explicitly not production-ready:

pip install duckdb --pre --upgrade

DuckDB now ships inside dbt

With dbt v2, released in September 2026, DuckDB is built in: no separate dbt-duckdb package, and the driver downloads automatically on first run. A minimal profile is all you need:

my_project:
  target: dev
  outputs:
    dev:
      type: duckdb
      path: ./warehouse.duckdb

This makes the combination of dbt and DuckDB the easiest way to learn analytics engineering on a laptop, and a fast way to test models before running them on Snowflake, BigQuery or Databricks. dbt v2 also supports DuckLake and Apache Iceberg catalogs through a catalogs.yml file, behind the use_catalogs_v2 flag.

What it means for you

• Analysts: keep using DuckDB for quick file analysis; 2.0 makes semi-structured JSON and cloud files much faster.
• Analytics engineers: try dbt v2 with DuckDB for local development and teaching. It is free and needs no warehouse account.
• Teams: server mode may replace "send me the Parquet file" workflows, but test it before relying on it.

Want to see SQL on your own files without installing anything? SQL Studio runs queries on CSV and Excel files right in your browser. For more on choosing tools, see Choosing the right database and The data stack in 2026.

Sources: the DuckDB blog posts "A Preview of DuckDB v2.0" (August 2026) and "DuckDB Now Ships inside dbt v2" (September 2026). Version 2.0 features may change before the final release.

Alessandro Ecclesie AgazziWritten by Alessandro Ecclesie Agazzi, analytics engineer in London. Updated 13 October 2026. About me