Skip to content

Dataset · openai

GDPval Benchmark for Real-World Work Tasks

openai/gdpval

220 expert tasks across 44 professions with prompts and files for evaluating AI on economically significant work.

What's inside

GDPval is a dataset for evaluating models on real economically valuable tasks. It contains 220 tasks across 44 professions; each consists of a text prompt and a set of auxiliary files. There is a warning about sensitive and political content, including NSFW topics.

What you can do with it

  • Evaluating LLMs on practical work tasks
  • Comparing models on the GDPval benchmark
  • Testing agentic systems on file handling
Size
up to 1K rows
Downloads this month
294.3K
Likes
579

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("openai/gdpval")