Comparison
Kaggle datasets: how to use them and where else to look
Kaggle datasets are free to download, but each one carries the licence its uploader picked; here is how to use them and what else exists.
Published
The short answer
Kaggle Datasets is Kaggle's public library of data uploaded by users and organisations. You search it on kaggle.com or with the Kaggle command line tool, and download through the site, the CLI or the kagglehub Python package. Every dataset carries a licence its uploader chose, from CC0 to non-commercial or unknown, so check it before any commercial use.
What Kaggle Datasets is
Kaggle Datasets is the part of Kaggle where people and organisations publish data for others to download, explore in notebooks and reuse.
How a dataset is organised
Each dataset lives at a URL made of the owner's username or organisation slug plus a dataset slug, and comes with a title, subtitle, description, file list and one licence. Kaggle's own API documentation describes the metadata file uploaders fill in, including that the licence list must contain exactly one entry.
What people want from it
People searching for Kaggle datasets usually want to find data on a topic, download it, or know whether they may use it at work. The last takes the most care: the licence is whatever the uploader selected, and the uploader may not own the data.
How to find and download Kaggle datasets
- 1
Search on the site or from the command line
On kaggle.com, use the Datasets search. From a terminal, the official CLI runs kaggle datasets list -s followed by your search term. Its filters include --file-type (csv, sqlite, json or bigQuery), --license (cc, gpl, odb or other), --sort-by (hottest, votes, updated or active) and --min-size and --max-size in bytes.
- 2
Download the files
kaggle datasets download owner/dataset-name fetches the whole dataset as a zip; -f picks a single file, -p sets the folder and --unzip extracts it and removes the archive. Adding a version number downloads that version.
- 3
Or load it in Python with kagglehub
kagglehub.dataset_download('owner/dataset-name') downloads the latest version into a local cache at ~/.cache/kagglehub/ unless you set KAGGLEHUB_CACHE. Kaggle says you only need to sign in for private datasets or public ones that ask for your consent first.
- 4
Read the licence before you use it
Open the dataset page and check the licence named there, then the description for any extra terms, such as a note that the data was scraped from another site.
Signing in to the CLI
The CLI signs in with an API token from your Kaggle settings, set as KAGGLE_API_TOKEN, saved to ~/.kaggle/access_token or obtained with kaggle auth login.
The licences on Kaggle datasets and what they allow
The Kaggle API accepts 28 licence identifiers for a dataset. They group into a handful of families, and the family decides whether a company can use the data in a product.
- Licence identifiers the Kaggle API accepts
- 28
- Licence per dataset, picked by the uploader
- 1
| Family | Kaggle identifiers | Commercial use | Watch for |
|---|---|---|---|
| Public domain | CC0-1.0, PDDL, US-Government-Works | Allowed; CC0's own summary says even for commercial purposes, without asking | Whether the uploader had the right to waive it |
| Attribution | CC-BY-3.0, CC-BY-4.0, CC-BY-3.0-IGO, ODC-BY-1.0, CDLA-Permissive-1.0 | Allowed with credit | Credit lines in anything you publish |
| Share-alike and copyleft | CC-BY-SA-3.0, CC-BY-SA-4.0, ODbL-1.0, DbCL-1.0, CDLA-Sharing-1.0, GPL-2.0, GPL-3.0, LGPL-3.0, AGPL-3.0, FDL-1.3 | Allowed, but derived works may have to carry the same licence | Whether your output counts as a derivative |
| Non-commercial | CC-BY-NC-4.0, CC-BY-NC-SA-4.0, CC-BY-NC-ND-4.0, CC-BY-NC-SA-3.0-IGO | Not allowed; the CC deed says you may not use the material for commercial purposes | Internal company use can still be commercial |
| No derivatives | CC-BY-ND-4.0 | Sharing allowed, changes not | Cleaning or relabelling may count as a change |
| Software and other | apache-2.0, EU-ODP-Legal-Notice | Depends on the text | Software licences fit data awkwardly |
| No clear grant | copyright-authors, other, unknown | Not established by the label | Ask the owner or treat it as all rights reserved |
Free dataset sources compared with their licences
Kaggle is one of several free places to get data. They differ in who uploads, how many datasets they say they hold and, most of all, how the licence reaches you. The last two rows are the paid routes people turn to when none of the free ones fit.
- Datasets on the Hugging Face Hub
- 1,124,548
- Datasets on data.gov
- 532,480
- Datasets on data.europa.eu, from 212 catalogues
- 1,817,651
| Source | What it is | Size it states | How the licence is shown | Commercial use |
|---|---|---|---|---|
| Kaggle Datasets | User and organisation uploads | Not published on the pages we checked | One licence per dataset, picked by the uploader from 28 options | Varies per dataset |
| Hugging Face Hub | Dataset repositories with a README card | 1,124,548 datasets | license field in the card metadata; other means a LICENSE file in the repo | Varies per dataset |
| Google Dataset Search | Search engine; hosts nothing | Millions of datasets, per Google's guide | Whatever the source page declares | Usage rights filter, based on provider values |
| data.gov | US federal open data catalogue run by GSA | 532,480 datasets | Federal data offered free and without restriction in most cases; non-federal data has its own terms | Mostly allowed; check each dataset's access and use section |
| data.europa.eu | Official EU data portal | 1,817,651 datasets from 212 catalogues in 36 countries | Per dataset, with a licensing assistant | Varies per dataset |
| UCI ML Repository | Academic collection, started in 1987 | Not published on the about page | Dataset pages we opened say CC BY 4.0 | Allowed with credit where CC BY 4.0 applies |
| Registry of Open Data on AWS | Directory of datasets stored on AWS | Not published on the page we read | Third parties under a variety of licences | Varies per dataset |
| Zenodo | CERN-run research repository | 50 GB per record by default | Open records must name a licence; closed and restricted records state terms in the description | Varies per record |
| Common Crawl | Web crawl archive | Not given on the terms page | Non-exclusive licence to the service; crawled pages may carry their owners' terms | Your responsibility; you indemnify Common Crawl, including for AI training |
| Licensed data marketplaces | Paid listings from registered providers | Varies | A licence agreed with the provider | Defined by the contract |
| Direct licence from the owner | A private deal, often arranged by a broker | One dataset | A data licence agreement you negotiate | Whatever you negotiate, including exclusivity |
When Kaggle datasets are not enough
Five signs you need another source
- The licence rules out your use. A non-commercial or unknown licence blocks a product launch, however good the data is.
- You need proof of provenance. Training data for an AI model is increasingly checked for where it came from and whether the people in it agreed; a public upload rarely documents that.
- Everyone else has it too. A competitor can download the same file, so it gives no edge. Exclusive data comes from a private licence.
- The data has to keep updating. Many uploads are one-off snapshots; a business feed needs a provider who commits to refreshes.
- What you need isn't public at all. Support tickets, code archives or transaction logs sit inside companies and only move through a negotiated deal.
What to use instead
Then the options are licensed data marketplaces or a direct deal with the company that holds the data, written up in a data licence agreement.
Where DataBroker fits, and where it doesn't
DataBroker is not a dataset library and holds no datasets. If you want a CSV to practise on tonight, Kaggle is faster and costs nothing, and we have nothing that competes with it.
What we offer
What we offer is a paid course and an app for people who broker company data: a lead finder that ranks companies likely to hold sellable data, value ranges for codebases, tasks and raw data, contract templates (mutual NDA, letter of intent, data license agreement, broker agreement) and a CRM.
The trade-off
That helps with the gap above: finding the company that holds private data and arranging a licence. It takes weeks, costs money, and deals can fall through. For open data, stay with the free sources.
More about kaggle datasets
Limits of this comparison
Sources
15 sources · checked Nov 9, 2026
The DataBroker team checked the facts on this page against 15 public sources, last on .
- Kaggle datasets pageKaggle presents them as open datasets · Checked
- Kaggle API docs: dataset metadataThe 28 licence identifiers, one licence per dataset, slug format · Checked
- Kaggle API docs: datasets commandsSearch, filter and download flags · Checked
- Kaggle API docs: READMEAuthentication options · Checked
- kagglehub READMEdataset_download, cache location, when sign-in is needed · Checked
- Creative Commons BY-NC 4.0 deedNon-commercial and attribution terms · Checked
- Creative Commons CC0 1.0 deedCommercial use without asking permission · Checked
- Hugging Face datasets page and licence docs (huggingface.co/docs/hub/repositories-licenses)Dataset count, license field, other licence handling · Checked
- Google News Initiative: Dataset Search quickstartMillions of datasets, filters based on provider values · Checked
- data.gov homepage and privacy policy (data.gov, data.gov/privacy-policy)Dataset count, GSA operator, federal data terms · Checked
- data.europa.eu homepageDataset, catalogue and country counts · Checked
- UCI ML Repository about page and Iris dataset pageFounded 1987, CC BY 4.0 licence statement · Checked
- Registry of Open Data on AWSThird-party datasets under a variety of licences · Checked
- Zenodo policies50 GB per record, access levels and licences, CERN operation · Checked
- Common Crawl terms of useLicence scope, third-party rights, indemnity · Checked
Kaggle datasets FAQ
Are Kaggle datasets free for commercial use?
Some are. Downloading costs nothing, but each dataset has its own licence. CC0 and CC BY datasets generally allow commercial use; non-commercial, unknown and copyright-authors datasets don't give you that right on their face.
How do I download a Kaggle dataset without the website?
Install the Kaggle CLI and run kaggle datasets download owner/dataset-name, or in Python call kagglehub.dataset_download with the same handle. Private datasets and those that ask for consent need your API token.
Can I search Kaggle datasets by licence?
From the CLI, yes: kaggle datasets list takes a --license filter with the values all, cc, gpl, odb or other. It groups licences broadly, so still read the exact licence on the dataset page.
What does an 'unknown' licence on Kaggle mean?
The uploader didn't state terms. That doesn't make the data free to use. Treat it as all rights reserved until the owner of the underlying data tells you otherwise.
Where can I get data Kaggle doesn't have?
Try the other free sources in the table first: Hugging Face, data.gov, data.europa.eu and Google Dataset Search. If the data is private, you need a licensed marketplace or a direct licence from the company that holds it.
Related
- Google Dataset Search
Google Dataset Search finds dataset pages that other sites publish; here is how it works and which tools also surface paid, licensed data.
- AI training data
AI training data licensing explained: which datasets labs want, publicly reported deal values, and the copyright and transparency rules that now apply.
Take the DataBroker course
The DataBroker course and tools: find companies with data to sell, price it, license it and close the deal.