UCI vs OpenML: Where Should You Get Tabular Machine Learning Datasets?

Answer in brief: Use the UCI Machine Learning Repository when you want the classic, widely cited benchmark datasets, such as Iris, Heart Disease, Wine Quality and Bank Marketing, with clear metadata. Use OpenML when you want thousands of datasets in one uniform format that load the same way from Python, R or Java and plug straight into benchmarking and AutoML. Both are free. Many UCI datasets are also mirrored on OpenML. This page is independent research from each project's official site and documentation.

Top pick for benchmarking at scale: OpenML. A single loader, standard metadata (FAIR and Croissant) and integrations with scikit-learn, mlr, Weka and PyTorch let you go from one dataset to a hundred in a single loop.

Head-to-head

	UCI ML Repository	OpenML
Run by	University of California, Irvine	Open-source community platform
Catalog	690 datasets	Thousands of uniformly formatted datasets
Signature datasets	Iris (150 rows × 4 features), Heart Disease (303 × 13), Wine Quality (about 4.9K rows), Bank Marketing (about 45.2K rows)	Many UCI classics mirrored, plus benchmark suites
Format	As donated, with standardized metadata pages	Stored in Parquet, with the same structure for every dataset
Access	Website download from each dataset page	Python, R and Java clients plus a REST API
Extras	Citations and variable descriptions; a new beta site is in testing	Tasks and shared evaluations for benchmarking; FAIR and Croissant metadata
Best for	Teaching, papers, single-dataset experiments	Large-scale benchmarking and AutoML
Cost	Free	Free

UCI: the classics, done right

Many machine learning courses still start at UCI, and for good reason. Its datasets are small and well described, so you can understand every column, and they are cited across decades of papers, which makes your results easy to compare with published work. New donations continue to arrive, and UCI is testing a redesigned beta site. The limits are scale and modality. These are small tabular and classical datasets, not foundation-model corpora.

OpenML: datasets built for experiments

OpenML treats a dataset as part of a reproducible experiment. Every dataset is stored in an efficient format with consistent metadata, so openml.datasets.get_dataset works the same way on any dataset ID. Tasks define the target and evaluation, which lets you compare methods on equal terms. Its documentation highlights FAIR principles and support for the Croissant metadata standard. Tabular data is its core strength; image support is limited.

When to pick which

• Teaching a class or writing a tutorial: UCI, for its familiar datasets and clean documentation.
• Benchmarking a new tabular method across many datasets: OpenML.
• AutoML research: OpenML, for its benchmark suites and its scikit-learn and mlr integrations.
• Citing a specific classic dataset in a paper: UCI's dataset page and citation.
• Need real-world tabular data plus free notebooks? See Hugging Face vs Kaggle.
• Need large public tables to query in SQL? See AWS Open Data vs BigQuery.

Pricing model

Both are free to browse, download and use through their APIs. Neither has a paid tier for dataset access. Check each dataset's license, which is set by its donor or uploader.

FAQ

Are UCI datasets available on OpenML?

Many are. OpenML mirrors a large number of UCI classics in its standard format, so you can load them with the same OpenML client.

How many datasets does UCI have?

The UCI site currently lists 690 datasets.

Can I use these datasets for deep learning?

They work well for tabular deep learning experiments, but they are far too small for pretraining foundation models. For that, see FineWeb vs Dolma.

Keep exploring

• The full ranked list of the best AI datasets
• UCI ML Repository review · OpenML review
• How to choose an AI dataset

Sources

• UCI Machine Learning Repository
• OpenML
• OpenML documentation

Named on this page as a provider? Join the network for a linked profile, Visit buttons and featured placement.