Kaggle Datasets review: the best AI datasets for learning and applied ML
Answer in brief: Kaggle is the best place to grab a dataset and start working on it right away. Its homepage lists 755K public datasets (checked November 2026), up from the 500,000+ it reported in February 2025. Public datasets are free to view, download and analyze. One click opens a notebook with free, quota-limited GPU and TPU time, and KaggleHub loads files straight into pandas, Polars or a Hugging Face Dataset. It's strongest for tabular, applied and competition data. For giant pretraining corpora, look elsewhere.
Best for: going from finding a dataset to running a notebook in minutes, with free compute.
Independent research from Kaggle's homepage, its official documentation (summarized by our research tool), the KaggleHub README and the live catalog, checked 2026-11-09. Not hands-on testing.
What it does for AI datasets
Kaggle datasets can be private or public, and anyone can view, download and analyze a public one. You can search the catalog and filter by size, file type, license and tag. The catalog also has curated groups, such as benchmark datasets. Each dataset connects to Kaggle Notebooks: versioned, cloud-hosted Python or R environments with the data already mounted.
The datasets worth grabbing here, by task
- Tabular and applied ML: the benchmark-datasets collection and thousands of domain datasets (transportation, retail, earth science and more).
- Competition data: datasets published by Kaggle itself, such as episode data from its AI battle challenges, refreshed daily while we checked.
- Code and agent evaluation: on the day we checked, a trending upload was FreshSWE-Py, a 5GB set of software-engineering tasks. As with any community upload, check its card and license.
- Dataset discovery data: Google's Dataset Search team publishes its metadata dataset on Kaggle.
For web-scale pretraining text, robotics or large speech corpora, the Hugging Face Hub has far more depth.
Who it's built for
Students, data scientists, Kaggle competitors and applied teams who want data and compute together without setting up cloud infrastructure.
Where it shines
- Scale of choice. 755K public datasets on the homepage count.
- Free accelerators. Kaggle's GPU guide gives a weekly quota of 30 hours, sometimes more depending on demand. The TPU guide allows up to 20 hours a week and up to 9 hours per session.
- KaggleHub. Load dataset files into pandas, Hugging Face or Polars objects. Inside a notebook, datasets are attached automatically and served from Kaggle's shared cache.
- Low friction. KaggleHub only needs authentication for private resources or public ones that require consent.
Honest drawbacks
- Quality and documentation vary between uploads.
- Each uploader chooses the license, so check it every time.
- Hardware changes: Kaggle retired its P100 GPUs on September 15, 2026 and pointed users to T4x2 notebooks, though some docs still mention the P100.
- Compute is quota-limited.
Pricing and licenses
Viewing, downloading and analyzing public datasets is free, and so are quota-limited notebook GPUs and TPUs. Licenses are set per dataset.
Best alternatives
- Hugging Face Hub: the deepest AI dataset catalog.
- UCI ML Repository: classic, well-cited benchmarks.
- OpenML: consistently formatted tabular datasets for benchmarking.
FAQ
How many datasets does Kaggle have?
Kaggle's homepage listed 755K public datasets when we checked on 2026-11-09. Kaggle doesn't date that figure.
Do I need an API key to download Kaggle datasets?
KaggleHub's README says you only need to authenticate for private resources or public ones that require consent.
How much free GPU time does Kaggle give?
Kaggle's GPU documentation states a weekly quota of 30 hours, sometimes more depending on demand, and the TPU guide allows up to 20 hours a week.
Are Kaggle datasets licensed for commercial use?
Only if the uploader's chosen license allows it. Check each dataset's license.
Sources
- Kaggle homepage (dataset count)
- Meet Kaggle (February 2025 figures)
- Kaggle Datasets
- Kaggle Datasets documentation
- Kaggle GPU usage docs
- Kaggle TPU docs
- Kaggle P100 sunset announcement
- KaggleHub README
← 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.