Artificial Analysis Intelligence Index

Artificial Analysis's own composite capability score, blending ten independently-run evaluations across agents, coding, general knowledge and scientific reasoning into one weighted number.

Also known as: Artificial Analysis Quality Index, AA Intelligence Index, AAII

unassessed

What it measures

The Artificial Analysis Intelligence Index — this repository's artificial_analysis_quality_index — is Artificial Analysis's composite capability score, built, in the publisher's own words, to give "a single score for tracking progress toward artificial general intelligence across mathematics, science, coding, and reasoning." It is not one test: version 4.3 blends ten independently-run evaluations across four weighted categories — Agents (30%: AA-Briefcase, GDPval-AA v2, AutomationBench-AA), Coding (20%: Terminal-Bench v4.0, SciCode), General (30%: AA-Omniscience, GDP.pdf, AA-LCR v1.1) and Scientific Reasoning (20%: Humanity's Last Exam, CritPt) — into one weighted number per model. The publisher's live site uses "Intelligence Index" as the current name for this score; this page documents it under this repository's id, which predates that branding.

Task format

A weighted blend of ten independently-scored evaluation tasks spanning agentic file/tool tasks, terminal-based coding tasks, open-answer knowledge and reasoning questions, and long-document and physics-reasoning problems; each component evaluation keeps its own response format and scoring rule before being combined.

Models reporting this benchmark

Data

This page as JSON · Edit on GitHub