Dataset · TabArena
BeyondArena large-scale tabular data benchmark
TabArena/BeyondArena
A large-scale dataset of 10 to 100 million rows with various tabular and text features for evaluating tabular models.
What's inside
BeyondArena is a unified benchmark for tabular classification and regression, covering tasks with independent and identically distributed (IID) data, time series, and group structures. The English-language corpus contains 10 to 100 million records from multiple disciplines, including text fields and high-dimensional categorical features. The dataset was introduced in a 2026 paper on evaluating the generalization capabilities of tabular foundation models.
What you can do with it
- Comprehensive evaluation and comparison of tabular foundation models
- Testing AutoML and gradient boosting algorithms on complex data with text and temporal structure
- Tasks
- Tabular classification, Tabular regression
- Languages
- English
- Size
- 10M–100M rows
- Downloads this month
- 19.8K
- Likes
- 12
How to load it
from datasets import load_dataset
ds = load_dataset("TabArena/BeyondArena")