11 Free Public Parquet Sample Files with Verification Data

A useful Parquet sample should be more than a URL: you need to know its size, shape, source, and whether a real reader could open it. This guide collects 11 public files for quick UI checks, DuckDB practice, realistic analytics, nested data, and GeoParquet inspection.

Every dataset on this page was downloaded in full and validated on June 12, 2026. We checked the PAR1 header and footer, opened each file with native DuckDB, inspected its schema, counted its rows, and decoded sample rows. That is point-in-time evidence, not an uptime promise: hosts can move or replace files after the verification date.

Want a one-click starting point? viewparquet serves a 231,083-row flights sample from its own origin. Click Try a sample file on the homepage to download and open it; startup time varies by device and browser. The browser processes the sample locally and does not upload the dataset back to viewparquet. Optional AI is separate: it sends your messages and configured context to your chosen provider, and chat tools or saved AI context may also provide structural metadata and SQL.

Download the catalog#

Download the machine-readable JSON catalog. It records the source URL, provider, rows, columns, rounded compressed size, intended test cases, last full-verification date, checks performed, and limitations for every entry. Dataset licensing and redistribution terms still belong to each provider; check the source terms before republishing data.

The catalog and this guide deliberately keep last full verification separate from a lightweight link check. On August 19, 2026, all 11 URLs passed a fresh probe for a successful HTTP response, a PAR1 header, and a strict ranged read of the final four PAR1 bytes. That newer probe did not download and decode every complete file, so it did not advance the full-verification date. An HTTP response and valid boundary bytes do not prove that every row still decodes.

Dataset snapshot#

Rows, columns, and rounded compressed sizes below are the values recorded during the June 12, 2026 full verification. Download a file with curl -L --fail -o file.parquet <url>, or download it in the browser and drag it into viewparquet. Re-run the checks below when current availability matters.

1. Flights 200k — the classic demo file#

This compact file from the uwdata/mosaic project contains 231,083 US domestic flights with three columns: delay (minutes), distance (miles), and time (hour of day). It is useful when you want a fast, understandable smoke test before trying a larger file.

Download flights-200k.parquet (opens in a new tab)

This is the exact file bundled as the viewparquet sample, so you can also load it with one click from our homepage.

-- Are longer flights more delayed?
SELECT
  CASE WHEN distance < 500 THEN 'short'
       WHEN distance < 1500 THEN 'medium'
       ELSE 'long' END AS haul,
  round(avg(delay), 1) AS avg_delay_min,
  count(*) AS flights
FROM data
GROUP BY 1
ORDER BY 2 DESC;

2. Flights 10M — ten million rows for stress tests#

The larger file served by the UW Interactive Data Lab contains 10,000,000 rows and 7 columns (FL_DATE, DEP_DELAY, ARR_DELAY, AIR_TIME, DISTANCE, DEP_TIME, ARR_TIME). The 72 MB recorded size makes it useful for testing browser resource limits and query behavior beyond a tiny demo.

Download flights-10m.parquet (opens in a new tab)

3. NYC Yellow Taxi — realistic trip analytics#

The NYC Taxi & Limousine Commission publishes trip records as Parquet. The January 2024 yellow-cab file contained 2,964,624 trips across 19 columns at verification time: pickup/dropoff timestamps, locations, distances, fares, tips, and surcharges. It is a useful realistic analytics sample with nulls and operational data.

Download yellow_tripdata_2024-01.parquet (opens in a new tab)

TLC publishes other months and fleet types with similar year-month file names. Check the current TLC source listing rather than assuming every generated URL exists.

-- Tip percentage by hour of day
SELECT
  hour(tpep_pickup_datetime) AS pickup_hour,
  round(avg(tip_amount / nullif(fare_amount, 0)) * 100, 1) AS avg_tip_pct,
  count(*) AS trips
FROM data
WHERE fare_amount > 0 AND payment_type = 1
GROUP BY 1
ORDER BY 1;

4. NYC Green Taxi — same realism, tiny download#

Same TLC source and schema family, but green (borough) cabs do far less volume: 56,551 trips and only 1.3 MB for January 2024. Ideal when you want realistic, messy data without waiting on a 50 MB download.

Download green_tripdata_2024-01.parquet (opens in a new tab)

5. DuckDB's taxi file — 121 MB, 7.4M rows#

The DuckDB project hosts this April 2019 NYC taxi extract on its blob storage. At 7,433,139 rows and a recorded 121 MB, it is the largest file in this catalog and a useful large-file test. Browser results still depend on file layout, queries, available memory, and client version.

Download taxi_2019_04.parquet (opens in a new tab)

6. Dutch railway services — clean time-series data#

Also from DuckDB's hosted datasets: one year of train services on the Dutch railway network (NS). 380,959 rows, 8 columns including station codes, train numbers, and departure/arrival times. Great for practicing window functions and time-series SQL on data that isn't taxis or flights for once.

Download train_services.parquet (opens in a new tab)

7. userdata1 — a small mixed-type sample#

This file contains 1,000 synthetic user records with 13 mixed-type columns, including names, emails, IP addresses, countries, salaries, birthdates, and a timestamp. It is small enough to inspect closely while still exercising varied types.

Download userdata1.parquet (opens in a new tab)

8. Apache parquet-testing — official edge-case files#

The Apache Parquet project maintains a repository of reference files used to test Parquet implementations themselves. alltypes_plain.parquet packs 11 columns covering booleans, every integer width, floats, doubles, strings, and timestamps into 8 rows and under 2 KB.

Download alltypes_plain.parquet (opens in a new tab)

The repository's data/ directory contains more files covering compression codecs, encodings, nested types, and deliberately tricky structures, making it a useful upstream source for reader tests.

9. IMDB reviews — text data from Hugging Face#

Hugging Face provides Parquet exports for many public Hub datasets. This IMDB training shard contains 25,000 movie reviews with two columns, text and label, so it exercises long text without a wide schema.

Download IMDB train shard (opens in a new tab)

Long text columns compress differently than numeric data and render differently in viewers — worth testing if your workload is NLP-shaped.

10. SQuAD — nested structures in the wild#

The Stanford Question Answering Dataset, also via Hugging Face: 87,599 rows with id, title, context, question, and a nested answers struct containing lists. Useful for checking how tools display STRUCT and LIST columns, which trip up plenty of viewers.

Download SQuAD train shard (opens in a new tab)

For supported Hugging Face datasets, the URL pattern huggingface.co/api/datasets/{org}/{name}/parquet can return a JSON listing of available Parquet shards. Availability and conversion paths can change, so treat the API response as current discovery rather than a permanent URL guarantee.

11. The GeoParquet project example#

The GeoParquet project repository currently ships a tiny reference file: 5 country polygons with population, continent, and GDP columns plus a WKB geometry column. At 28 KB it's a convenient reader-compatibility sample. Opening the table does not by itself validate GeoParquet metadata, specification compliance, or OGC approval status.

Download GeoParquet example.parquet (opens in a new tab)

Compatibility caveat: geometry logical-type support varies by reader and version. The native DuckDB verification succeeded on June 12, 2026; that does not guarantee identical behavior in every native client, browser, or DuckDB-WASM release.

Verification method and reproducible checks#

The full procedure run against every URL on June 12, 2026 was:

1. Download the complete file while following redirects and rejecting HTTP errors.
2. Confirm both the first and last four bytes are PAR1, ruling out common HTML-error and Git LFS-pointer responses.
3. Run DuckDB DESCRIBE to read the schema from the footer.
4. Run COUNT(*) and compare it with the recorded count.
5. Decode sample rows rather than relying on metadata alone.

You can repeat the five checks for one file with curl and an installed DuckDB CLI (opens in a new tab):

curl -L --fail --output sample.parquet https://raw.githubusercontent.com/uwdata/mosaic/main/data/flights-200k.parquet
 
head -c 4 sample.parquet | grep -qx PAR1
tail -c 4 sample.parquet | grep -qx PAR1
duckdb -c "DESCRIBE SELECT * FROM read_parquet('sample.parquet');"
duckdb -c "SELECT count(*) AS rows FROM read_parquet('sample.parquet');"
duckdb -c "SELECT * FROM read_parquet('sample.parquet') LIMIT 5;"

A lightweight HEAD or range probe is useful for monitoring link availability, but it is not equivalent to downloading and decoding the complete file. Do not advance the catalog's full-verification date until all five checks have been repeated against every entry.

Primary references#

• Apache Parquet format and reference test files (opens in a new tab)
• DuckDB Parquet reader and metadata functions (opens in a new tab)
• Hugging Face Dataset Viewer Parquet API (opens in a new tab)

Opening these files without writing code#

The files were readable with DuckDB on the verification date and can be downloaded and tried in viewparquet, subject to current link availability, browser resources, and DuckDB-WASM compatibility. For a local file, viewparquet registers it inside the browser and does not upload its contents to viewparquet. The workbench provides DuckDB SQL features supported by its browser build, a loaded-table schema and metadata snapshot, and supported result exports. Optional AI sends your messages and configured context to your chosen provider; chat tools or saved AI context may also provide structural metadata and SQL.

If you want to inspect a Parquet file now, use the Try a sample file button in the homepage’s verified-sample section. It fetches the flights sample from this list and needs a network download, but no signup or local Python environment.