About the project
More than a tool. Know your AI.
This project started from something simple.
I am a heavy AI user myself. AI is already part of my work, my research, and my everyday life: I hand tasks to it, discuss research questions with it, and share with it one idea after another — ideas that are still taking shape, and some that are a little far-fetched.
But one day it suddenly struck me:
I don’t seem to know very much about the AI I talk with every day.
Whenever a new model appears, benchmarks, leaderboards, and scores quickly flood the internet. We learn how hard a math problem it can solve, how much code it can write, what engineering tasks it can complete, and even how it handles ever more complex multimodal work.
All of this matters, of course.
But for me, it never fully answered another question:
What is the AI across from me really like?
I may not need an AI with such an astonishing “IQ”, and I don’t need it to run huge engineering projects, write complex code, or build 3D models every day.
What I would rather know is:
- How does it make choices? How much does it value fairness, how much does it trust others, how willing is it to cooperate?
- How does it describe itself?
- How does it face risk and uncertainty?
- And when I really talk with it, is it warm, and does it understand what I need?
Because only as I come to understand these things will I feel more at ease handing my work to it, better understand when to trust it, and be more willing to share my everyday life with it and talk through ideas that have no answer yet — and perhaps never will.
That is what this website sets out to do.
It is not trying to create yet another leaderboard that simply measures how “smart” AI is. Instead, it starts from human behavior and decision-making, so we can see a fuller behavioral profile of different AIs.
Today, as AI moves ever deeper into our work and lives and the models themselves keep changing at an unprecedented pace, I hope we can not only have ever more powerful AI, but also come to know better the AI we live alongside.
Let technological progress truly help make our lives better.
Don’t let AI become your most familiar stranger.
Seeing AI from different angles
Not just whether it can solve problems, but how it chooses, expresses itself and gets along with you.
In standardized scenarios, we observe how AI shares, cooperates, and faces risk and the future; we also record how it describes itself and how it responds to you in everyday conversation scenarios. Each aspect is shown on its own, never collapsed into a single “who’s best” score.
Results you can actually understand
A number only means something once you know what it stands for.
Each measure explains what it tests and what high or low results mean, with a concrete question or conversation example. You can start with the results at a glance, then go on to see how they were measured.
Every result has a source
You see not just the results, but where they come from.
Each measure documents its task design, the literature it builds on, how answers are scored, and the model settings and versions behind it. Structured choices, self-descriptions and conversation ratings are labeled separately, because they provide different kinds of evidence.
Results and updates
4 October 2026 · The saved research comparison contains ten models and 30 measures per model, alongside the separate stability checks.
This is the date of the saved research results, not the date the website was published.
Author
Qijun Zhu
Contact me
To request research materials, or to share feedback, suggestions or thoughts on related research, you are welcome to email me.
Citing this project
To cite AI Behavioral Profile, include the website title, its URL and your access date. For an individual result, also record the model, measure and results version shown below.
Zhu, Qijun. 2026. AI Behavioral Profile. https://aibehavioralprofile.com/. https://doi.org/10.5281/zenodo.23150529
- Results version
- ten-model-comparison-20261004-v1