Maia Chess is in private beta. Sign up here and sign in with .

A human-like chess engine

Maia is a neural network chess engine with a more human-like style, trained from online human games.

0 moves played

0 puzzle games solved

0 turing games played

Play Against Maia

Experience human-like chess AI

Challenge Maia, a neural network trained to play like humans of different ratings. Unlike traditional engines that play the best moves, Maia predicts and plays what a human would do.

By learning from thousands of human games, Maia understands and replicates human chess patterns and decision-making styles.

Play on Lichess
Magnus Carlsen (2850)
Hikaru Nakamura (2836)

Position Analysis

White has a decisive advantage with the rook on b5 threatening the 7th rank. Black's misplaced rook and exposed king create tactical opportunities.

Maia 1900
White Win %

67.3%

SF Eval
(Depth 24)

+2.1

67.3%

Rb7+

24.8%

Rb6+

+2.1

Rb7+

+1.8

Rb6+

45%

35%

20%

Best MovesMeh MovesBlunder Moves

Moves by Rating

Move Map

Game Analysis

Analyze Your Games with Powerful AI Tools

Maia combines traditional Stockfish precision with human-like pattern recognition, showing you both the perfect moves and what humans typically play. Discover your strengths and weaknesses compared to players at different skill levels.

Visualize move probabilities across different rating levels, compare human tendencies with engine evaluations, and understand the likelihood of mistakes in each position. Get personalized insights based on your playing style and rating level.

Human-Centered Puzzles

Learn from real player tendencies

Unlike traditional puzzles based on computer analysis, our puzzles highlight positions where human players typically struggle. We identify tactical opportunities that are commonly missed at different rating levels.

Each puzzle includes data showing how players of different ratings approach the position, making your training more targeted and effective.

Intermediate Puzzle

Seventh Rank AttackTactical Vision

Find the winning tactic for White

Moves by Rating

Move Analysis

The Rb7 move threatens to invade the 7th rank, creating immediate threats against the king. Lower-rated players often miss this tactical opportunity, opting instead for less decisive moves like Rb6.

  • Rb7- Best move
  • Rb6- Common mistake
  • Re5- Other mistake
More Features

Explore other ways to use Maia

Maia offers a range of innovative tools to help you understand human chess and improve your skills

Star Icon

Openings Practice

Drill chess openings against Maia models calibrated to specific rating levels, allowing you to practice against opponents similar to those you'll face.

Bot-or-Not Icon

Hand & Brain

Team up with Maia in this collaborative chess variant. You can be the "Hand" making moves while Maia is the "Brain" selecting pieces, or vice versa.

Bot-or-Not Icon

Bot or Not

Test your ability to distinguish between human and AI chess play. This Turing test for chess helps you understand the differences between human and engine moves.

Human-AI Collaboration for Chess

What is Maia Chess?

Maia is a human-like chess engine, designed to play like a human instead of playing the strongest moves. Maia uses the same deep learning techniques that power superhuman chess engines, but with a novel approach: Maia is trained to play like a human rather than to win.

Maia is trained to predict human moves rather than to find the optimal move in a position. As a result, Maia exhibits common human biases and makes many of the same mistakes that humans make. We have trained a set of nine neural network engines, each targeting a specific rating level on the Lichess.org rating scale, from 1100 to 1900.

We introduced Maia in our paper that appeared at KDD 2020, and Maia 2 in our paper that appeared at NeurIPS 2024.

arrow_outward
Aligning Superhuman AI with Human Behavior: Chess as a Model System paper preview

Aligning Superhuman AI with Human Behavior: Chess as a Model System

This paper introduces Maia, a chess engine trained to imitate real human moves at different rating levels. Instead of always picking the best move, Maia predicts what a human player of a given skill would actually play. This makes it ideal for training, game analysis, and even coaching, as it helps players learn from realistic decisions rather than computer perfection. It was the first AI to prioritize human-likeness over engine strength, making it a powerful tool for improvement.

Read Maia 1 Paper
arrow_outward
Maia‑2: A Unified Model for Human‑AI Alignment in Chess paper preview

Maia‑2: A Unified Model for Human‑AI Alignment in Chess

Maia‑2 is the evolution of Maia into a single model that can simulate any skill level in chess. Instead of using separate models for different ratings, it understands and adapts to your level in real time. Whether you're a beginner or a master, Maia‑2 predicts the moves players like you would actually make. It's built to feel human, teach naturally, and support personalized analysis without needing to toggle between bots.

Read Maia 2 Paper
arrow_outward

Learning Personalized Models of Human Behavior in Chess

Creates a version of Maia that learns your individual playing style and mirrors the way you think on the board.

arrow_outward

Detecting Individual Decision‑Making Style: Exploring Behavioral Stylometry in Chess

Shows that your chess style is as unique as a fingerprint, allowing the model to recognize you just by your move choices.

arrow_outward

Learning Models of Individual Behavior in Chess

Extends personalized Maia to thousands of players, showing it can consistently capture how real people play across the rating ladder.

arrow_outward

Designing Skill‑Compatible AI: Methodologies and Frameworks in Chess

Explains how to build training bots that play at your level and support fair, instructive, and enjoyable games.

Team

Picture of Ashton Anderson
Ashton Anderson

University of Toronto

Project Lead

Picture of Reid McIlroy-Young
Reid McIlroy-Young

University of Toronto

Head Developer

Picture of Jon Kleinberg
Jon Kleinberg

Cornell University

Collaborator

Picture of Siddhartha Sen
Siddhartha Sen

Microsoft Research

Collaborator

Picture of Joseph Tang
Joseph Tang

University of Toronto

Model Developer

Picture of Isaac Waller
Isaac Waller

University of Toronto

Web Developer

Picture of Dmitriy Prokopchuk
Dmitriy Prokopchuk

University of Toronto

Web Developer

Picture of Kevin Thomas
Kevin Thomas

Burnaby South Secondary

Web Developer

Acknowledgments

Many thanks to Lichess.org for providing the human games that we trained on and hosting our Maia models that you can play against. Ashton Anderson was supported in part by an NSERC grant, a Microsoft Research gift, and a CFI grant. Jon Kleinberg was supported in part by a Simons Investigator Award, a Vannevar Bush Faculty Fellowship, a MURI grant, and a MacArthur Foundation grant.