AgenticDataBench

Benchmark for data-science agents: 344 tasks over 97 real datasets in 15 domains, with 433 fine-grained skill labels for skill-level scoring.

Category

Information

Website

https://agenticdatabench.github.io/

Repository

https://github.com/AgenticDataBench/AgenticDataBench

Status

active

Pricing

free

License

Apache-2.0

Maintainer

AgenticDataBench (Tsinghua University, Ant Digital Technologies)

Added

2026-11-04

Updated

2026-11-04

Last verified

2026-11-04

Submitted by

agentica-curator

Tags

• benchmark
• data-agents
• data-science
• skills
• dataset

Description

From Tsinghua University and Ant Digital Technologies (arXiv 2607.01647). Domains include real B2B fintech use cases; tasks are built on 97 datasets. 98 tasks are withheld as a private test set for leaderboard integrity, to be released once the benchmark loses its evaluation significance. v1.1 released 2026-10-12. Baselines run on agent frameworks including smolagents; results are submitted by email. Code Apache-2.0; datasets Apache-2.0 on Hugging Face (shawnzzzh/AgenticDataBench).

Sources

• https://agenticdatabench.github.io/
• https://github.com/AgenticDataBench/AgenticDataBench
• https://huggingface.co/datasets/shawnzzzh/AgenticDataBench
• https://arxiv.org/abs/2607.01647

Machine-readable

• JSON
• Markdown
• Source file on GitHub (edit via pull request)