Maia Chess review: scores and evidence
Neural-network models that imitate human players by level (Maia 1100-1900, Maia 2 from 600 to 2600 Lichess), playable on Lichess and on a web platform with analysis, puzzles, opening drills and Bot-or-Not. No weakness diagnosis and no tracking; English interface.
Read on August 30, 2026 from the service’s public pages — every score cites the page read.
Overall score
3.0/ 10
Rank
39/ 128
Languages
1
Listed price
€0 (free, GPL-3.0 research project)
What the service is
Maia Chess is an academic research project run by the Computational Social Science Lab (CSSLab) at the University of Toronto, led by Ashton Anderson, with collaborators at Cornell (Jon Kleinberg) and Microsoft Research (Siddhartha Sen); maiachess.com also names the model developers and the web team. The project started with the paper "Aligning Superhuman AI with Human Behavior: Chess as a Model System" presented at KDD 2020, and its blog dates Maia’s public introduction to 24 August 2020; the web platform entered beta as v0 on 18 October 2024 and became v1 on 10 July 2025. Maia is a neural network trained on millions of human games to play like a player of a given level: the Maia 1 models target the 1100 to 1900 bands (playable as bots on Lichess), while Maia 2 covers 600 to 2600 with a single model. The site lets you play against Maia, analyse your games or positions with Stockfish and Maia in the browser (move probabilities by rating, Blunder Meter), solve puzzles, drill openings, play Hand & Brain and guess "Bot or Not". Everything is free, with no subscription; the model and platform code is public under GPL-3.0. The service produces no weakness diagnosis and no tracking over time, has no courses, coaches or mobile app, and exists in English only.
Activity history
- 20262026: the CSSLab/maia-chess GitHub repository announces a Maia-3 (Chessformer) paper for 2026; maiachess.com also lists it among its publications.
- 2025-0710 July 2025: "Introducing Maia Platform v1" — Maia 2 model, redesigned analysis page (Insight Panel, Moves by Rating, Blunder Meter, Move Map), PGN/FEN loading, an opening drill page, public profiles, leaderboards and mobile optimisation, in beta.
- 2024-1018 October 2024: "Beta Testing Maia Chess v0", first version of the platform to play and learn with Maia.
- 20242024: the paper "Maia-2: A Unified Model for Human-AI Alignment in Chess" is presented at NeurIPS 2024; the code is released under the MIT licence on GitHub (CSSLab/maia2).
- 2020-0824 August 2020: post "Introducing Maia: a Human-Like Chess Engine", presenting the model trained on millions of human games.
- 20202020: founding paper "Aligning Superhuman AI with Human Behavior: Chess as a Model System" at KDD 2020; the Maia 1100 to 1900 models are published under the GPL on GitHub, three of them playing as bots on Lichess.
The scores, evidence by evidence
Every score points to the page read and the fact found there. A fact documented nowhere is not credited: document it, then write to us.
APersonalized diagnosis
1.7A1 :2Importing your games (Chess.com, Lichess, PGN files)
github.com/csslab/maia-platform-frontend: "Lichess game import capability"; https://www.maiachess.com/: Sign in with Lichess
A2 :2Automatic engine analysis of every game
github.com/csslab/maia-platform-frontend: "Stockfish WASM + Maia" analysis of imported games; depth not documented
A3 :0Recurring mistakes identified across all your games
https://www.maiachess.com/: no aggregation
A4 :0Types of mistakes covered (tactics, strategy, endgame, time…)
https://www.maiachess.com/: no typology
A5 :0Written explanation of each mistake
https://www.maiachess.com/: "human insights" = move probabilities; no text
A6 :0Stated verification of the diagnoses (procedure, rates)
no mention on https://www.maiachess.com/
A7 :0Tracking each weakness over time
no history on https://www.maiachess.com/
A8 :0Comparison with players of the same rating
no comparison on https://www.maiachess.com/
BOpenings
2.2B1 :1Building a personal repertoire
https://www.maiachess.com/drills: "Learn and practice chess openings with Maia" — openings offered, not editable
B2 :2Training the repertoire (drills, games)
https://www.maiachess.com/drills: interactive opening practice against Maia
B3 :0Spotting where you left your repertoire in your games
no mention on https://www.maiachess.com/
B4 :0Verification of the lines taught (engine, practice)
no information on https://www.maiachess.com/
B5 :1Opening explorer and statistics
github.com/csslab/maia-platform-frontend: "Opening explorer"; database not specified
B6 :0Explanation of opening plans and ideas
no written teaching on https://www.maiachess.com/drills
CTargeted training
0.4C1 :0Exercises drawn from your own games
no exercise drawn from games on https://www.maiachess.com/
C2 :1Bank of tactical exercises (size, themes, rating)
https://www.maiachess.com/puzzles: puzzles present; volume and themes not published
C3 :0Endgame training
no endgame module on https://www.maiachess.com/
C4 :0Strategic and positional training
https://www.maiachess.com/: no positional exercise
C5 :0Training defense and alertness
https://www.maiachess.com/: nothing
C6 :0Training calculation and visualization
no mention on https://www.maiachess.com/
C7 :0Difficulty matched to your real level
no adaptive rating documented on https://www.maiachess.com/puzzles
C8 :0Spaced repetition (named algorithm)
https://www.maiachess.com/: no spaced repetition
C9 :0Personalized training plan
https://www.maiachess.com/: no plan
DPrice and value
10.0D1 :3What the free plan actually lets you do
https://www.maiachess.com/: entirely free
D2 :3Annual price of the full plan
https://www.maiachess.com/: €0
D3 :3Commercial transparency (price shown, trial, cancellation)
https://www.maiachess.com/: no price
D4 :3Number of paid tiers (how simple the offer is)
https://www.maiachess.com/: entirely free
EEvidence and transparency
5.8E1 :2Published evidence of effectiveness
https://www.maiachess.com/: KDD 2020, NeurIPS 2024, ICLR 2026 papers on prediction accuracy; no measure of progress
E2 :0Chess verification of the content (positions, lines)
no information on puzzle verification on https://www.maiachess.com/puzzles
E3 :3Transparency of the method (algorithms, code)
https://github.com/CSSLab/maia-chess + github.com/csslab/maia-platform-frontend: public code and models under GPL-3.0
E4 :2Marketing claims that can be checked (no overselling)
https://www.maiachess.com/: factual research wording
FContent and method
0.7F1 :0Courses organized by rating level
no courses on https://www.maiachess.com/
F2 :0Course authors named and titled
no course author; researchers named on https://www.maiachess.com/
F3 :0Content formats (text, video, interactive)
no content on https://www.maiachess.com/
F4 :0Published method of progression
no method on https://www.maiachess.com/
F5 :1How often the content is updated
https://github.com/CSSLab/maia-chess: Maia-3 published 2026; https://www.maiachess.com/ © 2026
GPlaying and practice
1.7G1 :0Playing humans online (pool, anti-cheating)
https://www.maiachess.com/: play against Maia or Hand & Brain with Maia; no humans
G2 :2Realistic artificial opponents
https://github.com/CSSLab/maia-chess: 1100-1900 models trained on human games; realistic thinking time not documented
G3 :0Replaying one specific position against the machine
replaying a position against Maia not observed on https://www.maiachess.com/
G4 :0Tournaments and organized competition
no tournament on https://www.maiachess.com/
HExperience and accessibility
1.3H1 :1Platforms available (web, iOS, Android, desktop)
https://www.maiachess.com/: web only (no store listing found via itunes.apple.com/search)
H2 :0User ratings and reviews (stores, press)
no store rating or third-party review observed for https://www.maiachess.com/
H3 :0Interface languages available
a single language observed — https://www.maiachess.com/: English only
H4 :0Accessibility (keyboard, screen reader)
no accessibility information on https://www.maiachess.com/
H5 :1Offline use and known incidents
https://www.maiachess.com/: depends on the server; no incident reported
IPreparation and study
0.0I1 :0Preparing for a named opponent
https://www.maiachess.com/: no opponent preparation
I2 :0Annotating and sharing games (studies)
https://www.maiachess.com/: analysis with no saving or sharing of annotations mentioned
I3 :0Entering a game played at a club (photo, scoresheet)
https://www.maiachess.com/: no game entry or scan mentioned
I4 :0Searching your games (filters, position)
https://www.maiachess.com/: counters (moves played, puzzles solved); no game list observed
JHuman coaching and community
2.5J1 :0Access to human coaches
no coach on https://www.maiachess.com/
J2 :1Size and activity of the community
discord.com/api/invites/Az93GqEAs7: "Maia Chess" 952 members, 102 online; daily messages not observed
J3 :0Tools for trainers, clubs and schools
no coach tool on https://www.maiachess.com/
J4 :2Engagement mechanics (streaks, virtual currency, reminders)
https://www.maiachess.com/: free, no paid mechanic
KPersonal data and interoperability
3.3K1 :0Exporting your data (games, repertoire)
no export mentioned on https://www.maiachess.com/
K2 :0Protection of personal data (GDPR, hosting)
https://www.maiachess.com/privacy: 404 — no policy
K3 :3Public API and source code
https://github.com/CSSLab/maia-chess: GPL-3.0
Also, outside the grid
Features found that no criterion measures. They do not count towards the score.
- Bot-or-Not (human or AI?)
- Hand & Brain variant with Maia
Sources consulted
- https://www.maiachess.com/
- https://www.maiachess.com/blog
- https://www.maiachess.com/blog/platform-v1
- https://github.com/CSSLab/maia-chess
- https://github.com/CSSLab/maia2
- https://github.com/csslab/maia-platform-frontend
- https://www.maiachess.com/drills
- https://www.maiachess.com/puzzles
- https://www.maiachess.com/privacy
Services with a similar profile
The three services whose category profile is closest — a distance over the eleven scores, computed, never chosen.