Skip to content

GAIA what it measures, how it is scored and who leads it

Agents and tools · Meta, Hugging Face and AutoGPT · introduced Nov 21, 2023 · Accuracy

About this benchmark

Task
Answer real-world questions that need browsing, tools, files and several steps.
Dataset
466 questions with unambiguous answers in three difficulty levels; most answers are held out.
Method
Exact match on a short final answer. People scored 92%, GPT-4 with plugins 15% at release.
Metric
Accuracy
Organization
Meta, Hugging Face and AutoGPT
Introduced
Nov 21, 2023

Versions

Newest first. New versions are added, never rewritten.

VersionDate
1Released with the paper.Nov 21, 20232 years ago

Sources: each benchmark's paper (arXiv) and its official site or leaderboard. Scores are the boards captured here every day, exactly as published.

New #1s by email

Saturdays, only in weeks when a leaderboard has a new #1.

Double opt-in. Unsubscribe any time.