{"slug": "maia-3-free-and-open-source", "title": "Maia-3: free and open source", "summary": "The University of Toronto's Computational Social Science Lab released and open-sourced Maia-3, a chess engine that predicts human moves rather than optimal ones, on Tuesday. The model achieves a 57.1% prediction accuracy on standard test positions, outperforming its predecessor Maia-2's 52.0% while using a fraction of the computational resources of comparable models. Maia-3 is now available for free on maiachess.com, where it can model players from 600 to 2600 Lichess rating, enabling human-like analysis, training, and gameplay for beginners and titled players alike.", "body_md": "# Introducing Maia-3: free and open source\n\n**A human-like chess engine open to all**\n\nToday we are releasing and open-sourcing **Maia-3**, our latest model of human behavior in chess! Maia's goal is to predict the human move, not necessarily the best move, in any chess position.\n\nThe Maia bots have been playing millions of games on Lichess for years. They're here to stay, but the Maia models that power them can do so much more. Maia-3 now models players from 600 to 2600, so it can power applications that apply to beginners and titled players alike. Maia-3 is the new state of the art, predicting the move 57.1% of the time on a standard test set of positions, compared to 52.0% for Maia-2 and 51.6% for Maia-1. ALLIE, another recent model, achieves 55.7% but needed an architecture almost 10x as large, making Maia-3 more efficient while still being more accurate.\n\n**For builders, download Maia-3**[here](https://huggingface.co/collections/UofTCSSLab/maia3), use the source code[here](https://github.com/CSSLab/maia3), and read the research paper[here](https://arxiv.org/pdf/2605.19091).**If you want to use Maia-3 to play, analyze, drill openings and endgames, do tactics, and more, go to**[www.maiachess.com](https://www.maiachess.com).** Join our**[Discord](https://discord.gg/hHb6gqFpxZ)to connect with our community. We'd love to hear what you think and see what you build using Maia-3!\n\nMaia-3 helps you enjoy and understand chess through a human lens. It lets you see not only the engine move, but which moves real players see, which mistakes they’re drawn toward, and how that changes the way you play, analyze, improve, and have fun with the game. Maia-3 unlocks lots of possibilities: human evaluation bars, human-like bots, new kinds of tactics puzzles, coaching aids, position difficulty measurement, and more. We can't wait to see what the community builds with it!\n\nFeatures\n\nMaia-3 has lots of benefits:\n\n- You can play against it for free on\n[www.maiachess.com](https://www.maiachess.com). It's a training partner that plays at your choice of skill level, is the most human-like engine out there, never gets tired, and is a low-stakes way to get some chess in. - Maia-3 now models players rated from 600 to 2600 (Lichess blitz ratings), so it spans 99% of the skill spectrum. Tools you build with it can apply to beginners or titled players.\n- Maia-3 powers the\n[www.maiachess.com](https://www.maiachess.com)platform, where you can do a new kind of analysis (combining the strength of Stockfish and the human side of Maia-3), play Hand and Brain, drill openings and endgames (pick some openings and endgames you want to practice, a skill level to train against, and go), try your hand at our chess Turing test called Bot-or-not (can you tell if a human or machine is playing?), follow tournament broadcasts and any Lichess game live, and more. For an example of Maia-3 powered analysis, see below. - It is designed to be interpretable, meaning it sees things on the chessboard similarly to how we see them. This makes it a lot easier to build educational and fun tools on top of Maia-3.\n\nChessformer, our new transformer architecture for chess modeling\n\nMaia-3 uses a completely new architecture called Chessformer. Concretely, Chessformer is an encoder-only transformer that treats the 64 board squares as tokens, pairs this square-token body with an attention-based “source-destination” policy head, and equips the trunk with Geometric Attention Bias (GAB), a novel dynamic positional-bias generator that adapts attention to the geometry of a chess position.\n\nChessformer is cool because it let us achieve the best human emulation performance, but we also integrated it into Leela Chess Zero and it resulted in a 100 Elo point gain for optimal play when search is turned off, plus it is interpretable by design, meaning it's a lot easier to figure out Maia-3's \"thoughts\". Prior to this, if you wanted to emulate human play you used one model (like Maia-2), but if you wanted raw engine strength you used another architecture (like Stockfish), but if you wanted something understandable you had to use a third strategy (like hand-crafted evaluations). Now you can just use Chessformer for everything. For those interested in all of the details, you can read our paper \"[Chessformer: A Unified Architecture for Chess Modeling](https://arxiv.org/pdf/2605.19091)\", which was just published in ICLR 2026 (a top machine learning conference).\n\nAnalyze games with Maia\n\nAs an example of what Maia-3 enables, try Maia-3 analysis on [www.maiachess.com](https://www.maiachess.com). You can analyze any game—your Lichess games, famous historical games, or any game or position you want—with a dual Maia-3 / Stockfish analysis view. By augmenting traditional engine analysis with Maia-3's human understanding, you get real-world context in every position. In Maia analysis, you see:\n\n**What Maia predicts** people at different rating levels would actually play, and with what probability**What Stockfish thinks** is objectively best**A \"Moves by Rating\" chart** showing how move choices shift across the skill spectrum, so you see what human progress looks like in every single position**Blunder detection** that understands which mistakes are natural and which are surprising for your level**Position danger analysis** that measures whether people at your level navigate this position well, or if they tend to make mistakes**A human evaluation bar** that tells you how good the position is from a human point of view, and at a given rating level. How likely is a 1200-rated player to win this position, or an 1800-rated player, or a 2500-rated player?\n\nThis gives you something we didn't have before: chess AI with human context. Your move might have been objectively second-best but completely natural for your rating. Another move might be the engine's top choice but almost no human below 2000 would find it. Some mistakes you make would even be played by GMs—but others you shouldn't be missing at your level (so you should probably work on those first). Maia-3's analysis lets you see chess through the lens of human play, and understand not just what's best, but what's realistic, where you're likely to stumble, what players slightly above your rating are doing differently, and what you need to change to get to the next level.\n\nFree and open source\n\nMaia Chess is an academic research project from the [University of Toronto CSSLab](http://csslab.cs.toronto.edu/). The models are completely free for you to download and use. The models are available [here](https://huggingface.co/collections/UofTCSSLab/maia3) and the code is available [here](https://github.com/CSSLab/maia3).\n\nBig thanks to Lichess.org\n\nThank you to Lichess for providing not only the data Maia-3 was trained on but for fostering an open and thriving global community of chess fans.\n\nJoin our community!\n\nJoin our [Discord](https://discord.gg/hHb6gqFpxZ) to connect with our community. We'd love to hear what you think and see what you build using Maia-3!\n\nTeam\n\nMaia-3 was made by:\n\n- Daniel Monroe\n- George Eilender\n- Philip Chalmers\n- Zhenwei Tang\n- Ashton Anderson\n\nHappy building!\n\n— Ashton Anderson and the Maia Chess team\n\n[Discuss this blog post in the forum](/ublog/vCPPRtX3/discuss)\n\n## You may also like\n\n[CM HGabor](/@/HGabor/blog/how-titled-players-lie-to-you/ickXiOem)\n\n## How titled players lie to you\n\nThis post is a word of warning for the average club player. As the chess world is becoming increasin…[thibault](/@/thibault/blog/how-i-started-building-lichess/JwtcE0KO)\n\n## How I started building Lichess\n\nI get this question sometimes. How did you decide to make a chess server? The truth is, I didn't.[ChessMonitor_Stats](/@/ChessMonitor_Stats/blog/where-do-grandmasters-play-chess-lichess-vs-chesscom/Zoi9GqPK)", "url": "https://wpnews.pro/news/maia-3-free-and-open-source", "canonical_source": "https://lichess.org/@/ashtonanderson/blog/introducing-maia-3-free-and-open-source/vCPPRtX3", "published_at": "2026-05-25 18:12:13+00:00", "updated_at": "2026-05-25 18:37:50.066750+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "neural-networks", "ai-research", "ai-products"], "entities": ["Maia-3", "Maia-2", "Maia-1", "ALLIE", "Lichess", "UofTCSSLab", "CSSLab", "maiachess.com"], "alternates": {"html": "https://wpnews.pro/news/maia-3-free-and-open-source", "markdown": "https://wpnews.pro/news/maia-3-free-and-open-source.md", "text": "https://wpnews.pro/news/maia-3-free-and-open-source.txt", "jsonld": "https://wpnews.pro/news/maia-3-free-and-open-source.jsonld"}}