Zenodo review: the best place for AI datasets behind research papers
Answer in brief: Zenodo is CERN's general-purpose open research repository, launched in May 2013 with OpenAIRE. It's where many academic AI datasets, benchmarks and model artifacts are published alongside papers, with citable records passed to DataCite. It puts no limits on format, size, access or license, supports embargoes and restricted records, and offers a rich API. It's free and trusted. Because it isn't built around ML, how well each dataset is documented and how easy it is to find varies.
Best for: finding the exact dataset a paper used, and publishing your own with a citable record.
Independent research from Zenodo's official About page, checked 2026-11-09. Not hands-on testing.
What it does for AI datasets
Zenodo accepts data, software and other research outputs from every discipline. You can keep content closed or restricted while research is ongoing, share protected links with reviewers, and embargo records so they open automatically when the paper is published. A web interface and an API let other tools use Zenodo as a backend. Citation information goes to DataCite and on to scholarly aggregators.
What to grab here, by task
- Benchmarks and evaluation sets released with academic papers.
- Domain datasets in science, medicine and the humanities that never reach ML hubs.
- Reproducibility artifacts: code snapshots, trained models and supplementary data for a specific paper version.
Who it's built for
Researchers reproducing or extending published work, and dataset authors who want permanent, citable hosting.
Where it shines
- Citable records through DataCite, which help authors get credit.
- Flexible access: open, restricted or embargoed.
- Open everything: the code is open source (built on InvenioRDM), metadata is CC0, and open content is reachable through open APIs.
- A trusted operator: CERN.
Honest drawbacks
- Not ML-specific: no dataset viewer, splits or ML loaders.
- Documentation quality depends on each depositor.
- Licenses vary by record, because Zenodo doesn't require any particular one.
Pricing and licenses
Free. Each depositor chooses the license; all metadata is CC0.
Best alternatives
- Hugging Face Hub: ML-native hosting with viewers and loaders.
- Google Dataset Search: find research datasets across repositories.
- Data.gov: official US government data.
FAQ
Who runs Zenodo?
CERN, which launched Zenodo in May 2013 as part of the OpenAIRE project.
Does Zenodo give datasets a DOI?
Zenodo makes records citable and passes citation information to DataCite.
Can I embargo a dataset on Zenodo?
Yes. Content can be embargoed and opened automatically when the associated paper is published.
Sources
- About Zenodo
- Zenodo policies
- Zenodo on GitHub
← Back to the ranked list of the best AI datasets
Named on this page as a provider? Join the network for a linked profile, Visit buttons and featured placement.