# What Is the Artificial Analysis Intelligence Index? v5 Scores, Explained

URL: https://whatisagenticai.net/blog/what-is-artificial-analysis-intelligence-index/
Published: 2026-10-28
Publisher: Little Tough LLC, maker of AI Agent Assistant: Agentic AI (https://apps.apple.com/us/app/ai-agent-assistant-agentic-ai/id6761501260)

The Artificial Analysis Intelligence Index v5 launched Oct 27, 2026. What the score means, the new private tests, and why Claude Opus 5.5 leads at 58.

Every time a new AI model comes out, one number shows up in the headlines: its score on the Artificial Analysis Intelligence Index. "Opus 5.5 hits 58." "Kimi drops." But what is that number, who makes it, and why did so many scores move this week?

The short version: the Artificial Analysis Intelligence Index is one of the most quoted independent AI leaderboards, and on October 27, 2026 it switched to a new version, v5, built mostly on tests that AI labs cannot see in advance. Here is what that means in plain English.

**TL;DR:** The Artificial Analysis Intelligence Index is a single score that combines many hard AI tests into one number, made by the independent benchmarking company Artificial Analysis. Version 5, announced October 27, 2026, now runs 60% of its tests on private data to stop models from "studying for the exam." Claude Opus 5.5 leads at 58, just ahead of GPT-6 Astra (57) and GPT-6.1 Sol (55). The gap between the best closed model and the best open-weights model grew from 3 points to 12.

## What is the Artificial Analysis Intelligence Index?

Artificial Analysis is an independent company that tests AI models and publishes the results. Its Intelligence Index takes a set of tough evaluations, covering things like coding, science, office work and general knowledge, runs every major model through them the same way, and blends the results into one score.

Think of it like a decathlon. One event does not tell you who the best athlete is, so you add up ten events. The Index does the same for AI models, which is why news stories use it as a quick way to say "this model is the smartest right now."

The number is not a percentage of human intelligence. It is a score on Artificial Analysis's own scale, and it changes meaning when the test set changes. That is exactly what just happened.

## What is new in Intelligence Index v5?

Version 5 is the biggest update since v4 arrived in January 2026. The headline change: 60% of the Index now comes from tests with private questions or hidden answers. Public tests leak onto the web, and models trained on web data can end up memorizing them. Private tests make that much harder.

Here is what changed, according to Artificial Analysis's announcement as reported by OfficeChai, AlphaSignal and TechApple:

| Test | Status in v5 | What it checks |
| --- | --- | --- |
| FrontierCode 2.0 | New | Private coding tasks from Cognition (132 tasks) in real open-source projects, graded on whether a human maintainer would merge the change |
| Terminal-Bench-Science | New | 70 science research tasks an AI agent must complete in a computer terminal |
| HLE-Diamond | Replaces Humanity's Last Exam | A refined 1,000-question set, half reasoning and half knowledge |
| AA-Omniscience v1.1 | Upgraded | Facts and avoiding made-up answers, with easy and public questions removed |
| CritPt v2 | Upgraded | A private 40-question set of research-level physics problems, with code tools allowed |
| GDPval-AA v2.1, SciCode, AA-LCR v1.1 | Dropped | Removed from the Index |

Tests carried over unchanged include AA-Briefcase (office work quality), AutomationBench-AA, GDP.pdf and Terminal-Bench 4.0. Artificial Analysis's announcement groups the tests into areas including Knowledge Work, Coding and Science at 25% each. One outlet, AlphaSignal, reported a different split that gives agent tasks more weight, so check the official methodology page if the exact weights matter to you.

Artificial Analysis also launched a companion, the **Progress Index v1**. Instead of resetting with every new version, it places every major language model released since November 2022 on one long-running scale, so you can see how far AI has come over time.

## Artificial Analysis Intelligence Index v5 leaderboard

These are the overall v5 scores reported by OfficeChai, AlphaSignal, TechApple and Startup Fortune. Most models were tested at their highest reasoning setting ("max").

| Model | Maker | v5 score | Note |
| --- | --- | --- | --- |
| Claude Opus 5.5 | Anthropic | 58 | 1st overall |
| GPT-6 Astra | OpenAI | 57 | Leads Terminal-Bench-Science and CritPt v2 |
| GPT-6.1 Sol | OpenAI | 55 |  |
| Gemini 4 Argon | Google | 50 | Best model outside Anthropic and OpenAI |
| Kimi K3.1 Pro | Moonshot AI | 46 | Top open-weights model (weights not yet released) |
| Claude Haiku 5.5 | Anthropic | 36 | Small, fast model |
| DeepSeek V4.1 Flash 1017 | DeepSeek | 32 | Tied for best downloadable open model |
| GLM-5.3 | Zhipu AI (Z.ai) | 32 | Tied for best downloadable open model |

Opus 5.5 also led on three single tests: FrontierCode 2.0 (46%), HLE-Diamond (68%) and AA-Omniscience v1.1. GPT-6 Astra won Terminal-Bench-Science at 63% to Opus's 59%. Smaller wins went elsewhere: Gemini 4 Argon topped AutomationBench-AA at 78%, and Meta's Muse Wonder Full topped GDP.pdf.

## Why did the scores change so much?

Two big shifts stand out.

**Open models fell behind.** Under the old v4.3 Index, the best open-weights model was within 3 points of the leader. Under v5 the gap is 12 points, and among models you can actually download today it is 26 points. Kimi K3.1 Pro, which we covered when it scored 55 and placed fourth on the old version ([What is Kimi K3.1 Pro?](https://whatisagenticai.net/blog/what-is-kimi-k3-1-pro/)), now sits at 46. Private tests seem to be harder to prepare for, and open models lost the most ground.

**Thinking costs more.** Models now write far more per task: the median output per Index task rose from about 35,000 to 59,000 tokens compared with v4.3, according to Artificial Analysis. Price also varies widely. Among the top ten, the cost to run one Index task ranges from $0.96 for GPT-6.1 Terra to $13.04 for Claude Fable 5.1, a spread of more than 13 times. If you want the cheap option, see [GPT-6.1 Sol vs Terra](https://whatisagenticai.net/blog/gpt-6-1-sol-vs-terra/).

## Artificial Analysis Intelligence Index vs other AI benchmarks

| Benchmark | Who runs it | What you get |
| --- | --- | --- |
| Artificial Analysis Intelligence Index | Artificial Analysis (independent) | One blended score from many hard tests, plus cost and speed data |
| Lab launch benchmarks | The company selling the model | Scores chosen by the maker, often on its own tests |
| Single benchmarks (like Terminal-Bench or Humanity's Last Exam) | Research groups | Deep view of one skill only |
| TasteVal | P-Zero Research | How well models plan and judge AI research experiments |

The Index is useful because it is independent and repeatable. It is not perfect. A higher score does not mean a model is better at your task, and a 1-point gap, like Opus 5.5's 58 versus Astra's 57, is too small to matter for most people.

## What this means for you

Leaderboards are a good signal for one thing: which few models are in the top group. Right now that group is Claude Opus 5.5, GPT-6 Astra and GPT-6.1 Sol, all within 3 points. Past that, the ranking shuffles every few weeks, and this week it shuffled just because the test changed.

For a normal person, the useful question is not "which model is number one" but "can I get the job done without becoming a benchmark expert." That is what [agentic AI](https://whatisagenticai.net/what-is-agentic-ai/) is for: agents that plan a task, do the steps and check their work, while the model choice happens in the background.

That is how **Agentic AI** works on iPhone and iPad. You pick a goal, and a team of ready-made expert agents running on frontier models gets to work. There is no account to create, and your first 3 tasks are free.

## Quick answers

**What is the Artificial Analysis Intelligence Index?** A single score from the independent company Artificial Analysis that combines many hard AI tests in coding, science, knowledge work and general skills, so models can be compared on one scale.

**What is Artificial Analysis Intelligence Index v5?** The new version announced October 27, 2026. It adds FrontierCode 2.0 and Terminal-Bench-Science, upgrades three tests, and runs 60% of the Index on private data.

**Which AI model is #1 on the Artificial Analysis Intelligence Index?** Claude Opus 5.5 at max reasoning, with a score of 58. GPT-6 Astra is second at 57 and GPT-6.1 Sol third at 55.

**What is the best open-weights model on the index?** Kimi K3.1 Pro at 46, though its weights are not released yet. Among downloadable models, DeepSeek V4.1 Flash 1017 and GLM-5.3 tie at 32.

**What does an Artificial Analysis Intelligence Index score mean?** It is a score on Artificial Analysis's own scale, not a percentage of human intelligence. Compare scores only within the same version: a v5 score and a v4.3 score are not the same test.

**Is the Artificial Analysis Intelligence Index reliable?** It is one of the most cited independent leaderboards, and moving to private tests makes cheating harder. Small gaps of 1 or 2 points still do not tell you which model is better for your task.

More terms like this in plain English: [the agentic AI glossary](https://whatisagenticai.net/agentic-ai-glossary/). Want to see how the top model works? Read [What is Claude Opus 5.5?](https://whatisagenticai.net/blog/what-is-claude-opus-5-5/)

**Sources:** [Artificial Analysis on X: Intelligence Index v5](https://x.com/ArtificialAnlys/status/2115265133467635877) · [Artificial Analysis on X: Progress Index v1](https://x.com/ArtificialAnlys/status/2115265195417403444) · [OfficeChai](https://officechai.com/ai/artificial-analysis-launches-intelligence-index-v5-with-majority-private-datasets-openai-models-gain-muse-wonder-kimi-drop/) · [AlphaSignal](https://alphasignal.ai/news/artificial-analysis-moves-intelligence-index-to-60-private-tests-to-stop-ai) · [TechApple](https://global.techapple.com/2026/10/artificial-analysis-intelligence-index-v5/) · [Startup Fortune](https://startupfortune.com/claude-opus-55-takes-top-spot-as-artificial-analysis-launches-intelligence-index-v5/)