Dataset · JobBench
Office task benchmark for AI agents across 35 professions
JobBench/job-bench
Set of real work tasks with documents and evaluation rubrics to check AI agent capabilities.
What's inside
Includes under 1k examples: prompts, reference files, weighted evaluation rubrics, and task descriptions for ~35 office professions. Divided into two parts: main (65 full tasks, some requiring finding needed files in a folder) and easy (63 simplified tasks without search).
What you can do with it
- Evaluating autonomous AI agent information retrieval and document handling skills
- Quick smoke testing and model comparison on the simplified easy split
- Size
- up to 1K rows
- Downloads this month
- 80.8K
- Likes
- 12
How to load it
from datasets import load_dataset
ds = load_dataset("JobBench/job-bench")