The open car database: every make and model, free to download

If you came here looking for a car database you can actually download, put in your app and not worry about, this is it: every make and model we know about, in one file, free.

It's 15,068 models across 934 makes, covering cars, motorcycles, mopeds, vans, trucks and buses, reconciled from official government vehicle data in 18 countries. It's licensed CC-BY 4.0, so you can use it commercially as long as you credit it.

One honest thing first, because "car database" means two very different things: this is the make and model database, the list of what vehicles exist. It does not have engine specs, trims, prices or photos per year. If that's what you need, the section on what it doesn't have (further down) tells you where to go instead. If you need a clean, complete list of makes and models for a dropdown, a lookup table or a data pipeline, read on.

Download it

The links below are the release this site serves, 2026.10.3 (published on October 10, 2026), so the counts on this page match the file you download. Newer releases appear on GitHub first. Pick the format you like:

Format	File	Size	Good for
CSV	vehicles.csv	1.2 MB	Excel, Google Sheets, any database import
JSON	vehicles.json	7.5 MB	Apps, the full nested record, with each country's rank
JSON (minified)	vehicles.min.json	0.8 MB	Shipping to a browser
Parquet	vehicles.parquet	0.3 MB	pandas, Polars, DuckDB, Spark
SQLite	catalog.sqlite	9.8 MB	SQL queries, per-country availability and popularity tables

Or from the command line:

curl -LO https://cdn.jsdelivr.net/gh/vehiclesdb/vehiclesdb@v2026.10.3/dist/vehicles.csv

Every file is also attached to the GitHub release, together with a manifest.json (version, counts per kind, sources) and ATTRIBUTION.md. No account, no form, no email.

Want it in Excel? Open vehicles.csv directly. Excel and Google Sheets both read it as is; pipe-separated columns like countries (de|gb|nl) split cleanly with "Text to columns" on |.

What's inside: the schema

One row per model. The CSV has these columns:

Column	What it holds	Example
kind	One of car, motorcycle, moped, van, truck, bus	car
make_slug	Stable id of the make	volkswagen
make_name	Display name	Volkswagen
model_slug	Stable id of the model within its make	golf
model_name	Display name	Golf
body_types	Body styles seen for the model, pipe-separated	hatchback|wagon
countries	Countries with official evidence of the model, ISO codes	ca|de|es|fi|gb|…|us
regions	Continents derived from the countries	as|eu|na|oc
global_popularity_decile	1 (top 10% most registered) to 10	2
mass_popularity_decile	1 to 10 within its kind, ranked by registrations summed across the registers: the "how many exist" decile	1
aliases	Other names the same model goes by	Rabbit
former_ids	Old ids that now point here	car/volkswagen/vw-golf|…

The canonical id of a model is kind/make_slug/model_slug, for example car/volkswagen/golf. Ids are stable: when a model is renamed or merged, the old id moves into former_ids instead of disappearing, so a foreign key you stored months ago still resolves.

The JSON carries the same records nested by make, plus each model's rank in every register that ranks it (country_ranks, e.g. {"gb": 2, "de": 1} for the Golf), and the SQLite file splits them into five tables: makes, models, availability (which source saw which model in which country), popularity (rank and decile per country) and meta.

Sample rows

Ten real rows from vehicles.csv, a few from each kind:

kind	make	model	body_types	countries	decile	aliases
car	Toyota	Corolla	hatchback, wagon	all 18	1	カローラ, Corolla Touring Sports
car	Tesla	Model 3	sedan	17 of 18	1
car	Dacia	Sandero	hatchback	ch, de, es, fi, gb, ie, il, lu, nl, no, ua	2	Renault Sandero
car	Volkswagen	Golf	hatchback, wagon	17 of 18	2	Rabbit, Golf Variant
car	Škoda	105	sedan	ch, fi, gb, nl	8
motorcycle	Yamaha	MT-07		au, ch, es, fi, gb, il, lu, nl, nz, th, ua	1
motorcycle	Honda	CBR600RR		ch, fi, il, lu, nl, th, ua	3
moped	Piaggio	APE50		ch, fi, gb, lu, nl, nz	7
van	Ford	Transit		au, ch, de, es, fi, gb, il, lu, nl, no	1
truck	Scania	114		fi, nl, no	7

A few things worth noticing. The Golf's US alias, Rabbit, is in there. The Škoda 105, a 1970s and 80s Czech saloon, is still in the Finnish, British and Dutch registers (and the Swiss canton of Thurgau's), so it's in the database. And the Dacia Sandero lists "Renault Sandero" as an alias, because that's what it's called in some markets.

Counts by kind

Kind	Models	Makes
Cars	5,541	315
Motorcycles	6,033	266
Mopeds	1,399	315
Vans	739	130
Trucks	947	95
Buses	409	94
Total	15,068	934

A make can appear under several kinds (Honda makes cars, motorcycles, mopeds and vans), which is why the per-kind make counts add up to more than 934.

Use it as a car dataset in pandas

If you're here for a car dataset for a notebook, the Parquet file is the quickest way in:

import pandas as pd

url = "https://cdn.jsdelivr.net/gh/vehiclesdb/vehiclesdb@v2026.10.3/dist/vehicles.parquet"
cars = pd.read_parquet(url)

cars["kind"].value_counts()
# motorcycle    6033
# car           5541
# moped         1399
# ...

# Every car model registered in Germany
in_germany = cars[(cars["kind"] == "car") & cars["countries"].str.contains("de")]

Or with SQL, straight against the SQLite file:

-- The ten most popular car models in the Netherlands
SELECT m.name, m.make_id, p.rank
FROM popularity p
JOIN models m ON m.id = p.model_id AND m.kind = p.kind
WHERE p.kind = 'car' AND p.country = 'nl'
ORDER BY p.rank
LIMIT 10;

A fair warning for machine learning projects: this is an identity table. It has no prices, mileage, fuel consumption or images, which is what most "car dataset" projects on Kaggle are about (price prediction, image classification). Where it shines is as the join table: normalise the make and model columns of a messy dataset to these ids, and suddenly two datasets that spelled "VW Golf" and "VOLKSWAGEN GOLF" differently line up.

Motorcycles, trucks and buses too

Most car databases stop at cars. This one has the same structure for every kind of vehicle the registers record. That makes it a motorcycle database (6,033 models) and a truck database (947 models) as well, with the same ids, the same columns and the same licence. Filter on kind and you're done.

Coverage is honest about where it comes from: motorcycle evidence comes from the registers that record motorcycles by model (Spain, Finland, the UK, Luxembourg, the Netherlands, New Zealand, Thailand and Ukraine), so a motorcycle sold only in the US may be missing.

What it doesn't have, and where to get it

So you don't waste an afternoon finding out:

• Engine specs, trims and dimensions per model year. Spec encyclopedias like auto-data.net and commercial feeds like CarAPI sell those.
• Prices and valuations. That's the business of KBB, Edmunds and the market-data vendors.
• Used-car listings (the "used car dataset" people train price models on). Kaggle has several, scraped from classifieds.
• VIN decoding. For US vehicles, the free NHTSA vPIC API decodes VINs.

What the paid side of VehiclesDB adds on top of this file is depth on the same ids: production runs, variants and reference links per model through the API, plus a resolver for messy strings.

Where the data comes from

Every model is built from official government data: vehicle registration data from the Netherlands (RDW), the United Kingdom (DfT), Spain (DGT), Germany (KBA), Finland (Traficom), New Zealand (NZTA), Luxembourg (SNCA), Ireland (CSO), Ukraine (the Ministry of Internal Affairs), Malaysia (JPJ), Thailand (DLT) and Argentina (DNRPA), plus the official fuel economy listings of the United States and Canada. A model only ships when at least two independent sources agree it exists, or when one source shows a fleet count no typo could produce. We publish the measured error rate instead of asking you to trust us.

Licence: CC-BY 4.0 and what attribution looks like

You can use the data for anything, including commercial products, as long as you give visible credit with a link. The simplest version:

<a href="https://vehiclesdb.com">Vehicle data by VehiclesDB</a>

The attribution page has ready-made snippets for apps, READMEs and academic citations (there's a Zenodo DOI). If you'd rather not show a credit, a paid plan from Scale up includes the waiver; the register notices in ATTRIBUTION.md apply either way.

Versions and updates

The dataset is versioned by date (YYYY.MM.PATCH) and rebuilt from the registers on a monthly schedule. New releases appear on GitHub; the version you're using is in manifest.json and in every API response, so you always know which snapshot an answer came from. Ids don't break between versions; that's what former_ids is for.

When the file isn't enough

Three things people usually need next:

• Your own data is messy. If you have a column of free-text vehicle names ("vw golf", "MERCEDES-BENZ | 300 TDT"), the resolve API maps each one to an id in this file, with a confidence and an audit trail. Every account gets 100 free credits, no card.
• You need it in a form. The make and model dropdown is this data behind one script tag, and it costs 0 credits.
• You want to browse first. Everything in the file is on the site, from the list of makes to every model page.