▶AI Research Directory
ds

DSBench

Data science agent benchmark

Open-sourceResearch Agents & AI ScientistsLLM as Analyst

AT A GLANCE

Intended use: Evaluating DS-agent capability on standardized tasks.

Pricing model: Open-source. Open-source availability does not establish free hosted usage, model access, or compute.

Public-source review: . Documentation checked; product performance was not tested.

LIMITATIONS

  • DSBench is a different project from THUDM's DataSciBench.
  • Repository code is MIT-licensed; the dataset has separate noncommercial-use restrictions.

FEATURES

  • ▸ Includes separate data-analysis and data-modeling task sets.
  • ▸ Evaluates solutions generated from task instructions and supplied data files.

What it is

A benchmark for data-science agents using realistic data-analysis and data-modeling tasks drawn from ModelOff and Kaggle.

RELATED · SAME CATEGORY