BigQuery public datasets review: the best AI datasets you can query in SQL

Answer in brief: BigQuery public datasets let you explore and extract large datasets with SQL without downloading anything. Google pays for the storage, you pay only for queries, and the first 1TB processed each month is free, so you can start without enabling billing. It's ideal for building training or feature sets from structured data such as GitHub activity, Wikipedia revision history and NOAA weather. There's no SLA, and you need a Google Cloud project.

Best for: SQL-first teams pulling ML features and training slices out of large structured datasets.

Independent research from Google Cloud's official BigQuery documentation, checked 2026-11-09. Not hands-on testing.

What it does for AI datasets

Public datasets are stored in BigQuery and shared through the Google Cloud Public Dataset Program. You query them with GoogleSQL or legacy SQL using fully qualified names, for example bigquery-public-data.bbc_news.fulltext. Access is through the Cloud console, the bq command-line tool, the REST API or client libraries for Java, .NET and Python. You can also browse and subscribe to public datasets through BigQuery sharing (formerly Analytics Hub) and the datasets section of Cloud Marketplace.

The datasets worth grabbing here, by task

• Text and NLP: sample tables with the complete Wikipedia revision history up to April 2010, English trigrams from works published 1520–2008, and a Shakespeare word index.
• Code and developer activity: GitHub timeline tables of pull requests and comments, created in 2012.
• Tabular and time series: NOAA weather (GSOD, 1929–2010) and US births 1969–2008.
• Medical imaging: Google Cloud also documents the NIH chest x-ray and The Cancer Imaging Archive datasets.

Who it's built for

Analysts and ML engineers who live in SQL, and teams already on Google Cloud.

Where it shines

• No download step. Query in place and export only what you need.
• Generous free tier. The first 1TB of queries per month is free, and Google says new customers get $300 in free credits.
• Many ways in: console, CLI, API and client libraries.

Honest drawbacks

• Requires a Google Cloud project, and queries beyond the free tier are billed.
• No SLA for the Public Dataset Program.
• Sample tables live in the US multi-region, so you can't join them with tables in other regions.
• Some samples are old (Wikipedia stops in 2010). Check each table's Last modified date.
• Not accessible by default from inside a VPC Service Controls perimeter.

Pricing and licenses

Google pays for storage. You pay for queries after the first free 1TB each month. Licenses and terms are set per dataset.

Best alternatives

• Registry of Open Data on AWS: cloud-scale scientific and web data.
• Azure Open Datasets: curated enrichment data.
• Kaggle: free notebooks for smaller datasets.

FAQ

Are BigQuery public datasets free?

Storage is free because Google pays for it. Queries are free up to the first 1TB processed each month and billed after that.

Do I need billing enabled?

Not to start. Google says you can query within the free tier without enabling billing, but you need a Google Cloud project.

Is there an SLA?

No. Google states that the Public Dataset Program has no service-level agreement.

Sources

• BigQuery public datasets documentation
• BigQuery pricing
• Google Cloud Public Dataset Program

← Back to the ranked list of the best AI datasets

Named on this page? Join the network for a linked profile, Visit buttons and featured placement.