DuckDB vs BigQuery

Quick verdict

DuckDB is an in-process analytical (OLAP) SQL database. It runs inside a host program with no server, uses a columnar, vectorised engine, and queries Parquet, CSV and JSON files directly on local disk or object storage, including Google Cloud Storage through its S3-compatible interface. It is free under the MIT License. duckdb.org lists 1.5.6 (28 September 2026) as the latest stable release and 1.4.5 as the current LTS release, with 2.0.0 scheduled for 21 October 2026.

BigQuery is Google Cloud's serverless data warehouse. Google describes a "serverless architecture" with "zero infrastructure management" and separate compute and storage layers. It supports open table formats such as Apache Iceberg, Delta and Apache Hudi through BigLake, and BigQuery Omni can query data held in AWS and Azure. You do not size or start clusters: you send SQL (GoogleSQL) and pay either for the bytes each query processes or for reserved slot capacity.

For DuckDB against another warehouse with a different billing model (credits for running virtual warehouses), see DuckDB vs Snowflake. This page focuses on what is specific to BigQuery: per-byte billing, the sandbox and reading BigQuery data from DuckDB.

Side by side

Key differences

Paying per byte scanned versus free local compute

BigQuery's on-demand model charges for the number of bytes each query processes, with the first 1 TiB per month free. Google's pricing page adds details that matter for exploratory work: charges are rounded up to the nearest MB, with a minimum of 10 MB per table referenced and 10 MB per query; queries that error or that return cached results are not charged; and cancelling a running query may still incur charges up to the full cost. Google's cost guide also warns that on a non-clustered table, adding LIMIT does not reduce the data read. A SELECT * over a wide table therefore costs the same whether you look at ten rows or ten million.

BigQuery gives you tools to control this: a dry run returns the estimated bytes without charge, and a maximum bytes billed setting makes a query fail, without charge, if its estimate is over the limit. The alternative is capacity pricing, where you pay for slots (virtual CPUs) over time in the Standard, Enterprise or Enterprise Plus edition, which suits steady, heavy use better than bursts.

With DuckDB the compute is the machine you already have, so running the same query fifty times while you refine it costs nothing extra. In our view this is the main practical reason people reach for DuckDB alongside BigQuery: iterative exploration of a dataset that fits on a laptop, without watching the bytes counter.

The BigQuery sandbox versus DuckDB as a learning tool

Both are free ways to start. The BigQuery sandbox needs no credit card or billing account. Google lists its limits: a lifetime 10 GiB of storage (not refunded when you delete data), 1 TiB of processed query data per month (the same as the free tier), and a default 60-day expiry on all tables, views and partitions. It does not support streaming data, DML statements (INSERT, UPDATE, DELETE, MERGE) or the BigQuery Data Transfer Service.

DuckDB has no such limits beyond your disk and memory, supports DML, and works offline. The sandbox, on the other hand, gives you BigQuery's public datasets, its console and GoogleSQL, which matters if the job you are preparing for uses BigQuery. For learning SQL itself, either works; for learning BigQuery, use the sandbox. To practise, see the SQL exercises.

Reading BigQuery data from DuckDB

DuckDB can read BigQuery tables through the bigquery community extension. It is community-maintained (by hafenkran), MIT licensed, and not an official DuckDB or Google project. It reads native tables through the BigQuery Storage Read API with projection and filter pushdown, can run GoogleSQL in BigQuery with bigquery_query(), and supports Linux, macOS and Windows (amd64) builds but not WebAssembly. Authentication can use Google Application Default Credentials.

-- DuckDB: read a BigQuery table, then keep a local Parquet extract
INSTALL bigquery FROM community;
LOAD bigquery;
ATTACH 'project=my_gcp_project' AS bq (TYPE bigquery, READ_ONLY);

COPY (
    SELECT order_id, region, amount
    FROM bq.sales.orders
    WHERE order_date >= DATE '2026-09-01'
) TO 'orders_sept.parquet' (FORMAT parquet);

Reads through the Storage Read API are billed by BigQuery, not by DuckDB: Google lists a free allowance of 300 TiB of bytes read per month per billing account, after which reads are charged per TiB, and network egress charges apply to data leaving Google Cloud, for example to your laptop. Queries passed through bigquery_query() run in BigQuery and are billed like any other query. The extension's README itself reminds users that BigQuery is a paid service and suggests budget alerts.

The other direction uses open formats. DuckDB writes Parquet that BigQuery can load or query from Cloud Storage, and DuckDB's iceberg extension documents Google Cloud BigLake among the Iceberg REST catalogs it can attach.

Serverless at scale versus a single machine

"Serverless" means different things here. BigQuery is serverless in the cloud sense: Google allocates compute for each query from a shared pool (or your reserved slots), so the size of the data is not limited by any one machine you own, and many users can query at once with Google Cloud IAM controlling access. DuckDB is serverless in the embedded sense: there is no server process at all, and the limit is the CPU, memory and disk of the machine running it, with documented spilling to disk for large joins, aggregations and sorts.

Editorially, DuckDB is usually enough when one person or process analyses data that fits on one machine and access can be controlled through files or buckets. BigQuery earns its cost when data is shared across a team, too large for one machine, or already flowing into Google Cloud from other services.

Pricing and licensing

DuckDB is free and open source under the MIT License, with no paid edition. Costs are your own hardware plus any cloud storage reads and egress. Extended LTS support is offered commercially by DuckDB Labs; prices are not published. MotherDuck, a separate company, sells a hosted service built on DuckDB; see DuckDB vs Snowflake for its plans.

BigQuery has two compute pricing models. On-demand is listed on Google's pricing page in October 2026 at USD 6.25 per TiB processed in US regions such as Iowa (us-central1), with the first 1 TiB per month free per billing account; prices differ by location. Capacity pricing bills slot-hours in the Standard, Enterprise or Enterprise Plus edition, with optional one- or three-year commitments in Enterprise and Enterprise Plus. Storage is billed separately per GiB (logical or physical, active or long-term), with the first 10 GiB per month free. The Storage Read API has its own per-TiB charge after a free 300 TiB per month, and network egress is charged separately. The sandbox is free without a billing account, within the limits above.

Prices exclude tax and change; use Google's pricing calculator for estimates.

Pricing checked on the vendors' official pages on 7 October 2026. Prices change; confirm before buying.

Where each one leads

Limitations

When to choose each

When neither is right

• You are choosing between cloud warehouses: see Snowflake vs BigQuery and BigQuery vs Redshift.
• You need a self-hostable server for real-time analytics: see ClickHouse vs BigQuery.
• Your analytics are modest and your data sits in an application database: PostgreSQL may be enough; see BigQuery vs PostgreSQL and DuckDB vs PostgreSQL.
• You need a local database for an application rather than analysis: see DuckDB vs SQLite.

Final recommendation

Frequently asked questions

Sources

• DuckDB: Release calendar
• DuckDB community extension: bigquery
• duckdb-bigquery repository (README and licence)
• DuckDB: Iceberg REST catalogs
• BigQuery: Introduction
• BigQuery pricing
• BigQuery: Sandbox
• BigQuery: Estimate and control costs

Checked October 2026.

How we research comparisons: our editorial method.