Benchmarks · 2024

AA Intelligence Index: Artificial Analysis Intelligence Index

Name: AA Intelligence Index: Artificial Analysis Intelligence Index
Creator: Artificial Analysis
Published: 2024
Keywords: AA Intelligence Index, AI benchmark, text model evaluation, Artificial Analysis

Artificial Analysis composite score that blends a dozen reasoning, coding, and math benchmarks into a single number.

Open Dataset

Models Tested

169

Top Score

57.6

Published

2024

Source

Artificial Analysis

Tools for AI Builders—All the hardware and tools you need to run private AI at your office or at home.Shop Now

How It Works

The Intelligence Index is Artificial Analysis’s way of giving every LLM a single comparable score across reasoning, coding, math, and long-context ability. They run a dozen public benchmarks under the same prompt and scoring settings, normalize each one to a 0–100 range, and average them with fixed weights. It is most useful as a quick way to spot the top tier without studying every sub-benchmark.

Artificial Analysis runs each model on the underlying evaluations using a shared prompting harness, normalizes each per-benchmark score, and aggregates them with published weights. The composite is recomputed whenever a new model lands or a benchmark is added.

Dataset size

Weighted aggregate over the Artificial Analysis evaluation suite, including MMLU-Pro, GPQA Diamond, HLE, MATH-500, LiveCodeBench, SciCode, IFBench, and AA-LCR.

Mean score

19.9

Median score

16.8

Open / Closed

73 / 96

Top Scorers

#	Model	Lab	Source	Score
01	Claude Opus 5.5	Anthropic	Closed	57.6
02	Claude Sonnet 5.5	Anthropic	Closed	56.0
03	Gemini 4 Argon	Google	Closed	52.6
04	GPT-6.1 Sol	OpenAI	Closed	51.8
05	Claude Fable 5	Anthropic	Closed	49.6
06	GPT-6 Sol	OpenAI	Closed	47.6
07	GPT-5.6 Sol	OpenAI	Closed	47.0
08	Qwen3.8-Max	Alibaba	Open	45.4
AdTools for AI BuildersAll the hardware and tools you need to run private AI at your office or at home.Shop Now
09	Kimi K3	Moonshot AI	Open	43.6
10	GPT-5.6 Terra	OpenAI	Closed	42.1
11	Claude Opus 4.8	Anthropic	Closed	41.8
12	Ling-3.1-flash	Ant Group	Closed	41.1
13	Claude Opus 4.7 Thinking	Anthropic	Closed	40.7
14	Claude Opus 4.7	Anthropic	Closed	40.7
15	DeepSeek-V4.1-Flash	deepseek-ai	Open	39.5

Score Distribution

Open vs Closed Source

Gap on AA Intelligence Index:+12.2pts closed leads

Top Open-Source Models

1Qwen3.8-Max45.4
2Kimi K343.6
3DeepSeek-V4.1-Flash39.5

Top Closed-Source Models

1Claude Opus 5.557.6
2Claude Sonnet 5.556
3Gemini 4 Argon

Score vs Parameter Count

Average Score by Lab

Anthropic
29.0n = 21
Xiaomi
26.6n = 3
MiniMax
24.9n = 3
Moonshot AI
24.3n = 7

Most Correlated Benchmarks

HLE
+0.94n = 165
Terminal Bench Hard
+0.93n = 129
SciCode
+0.86n = 64
LiveCodeBench
+0.86n = 86
SWE-Pro

What It Captures Well

A single comparable number across most modern LLMs.
Run consistently by one team using one harness, so cross-model comparisons are fair.
Updated continuously as benchmarks and models change.

Where It Falls Short

Composite scores hide which specific skills a model is good or bad at.
Weighting choices are opinionated and can favor certain skill profiles.
Does not capture agent ability or production reliability.

Frequently Asked Questions

How is this different from our averageScore?

Our average pulls in every benchmark we track, including Arena scores and HF leaderboards. The AA Intelligence Index is Artificial Analysis’s own composite over their evaluation suite. Use it as a second opinion, not a replacement.

Can I see the underlying scores?

Yes. Every benchmark inside the Intelligence Index has its own deep-dive page, so you can drill into MMLU-Pro, GPQA Diamond, LiveCodeBench, and the others independently.