Play backgammon online. Understand every move.
Ranked matches with an Elo rating, live games with friends, a bot ready at any hour. Behind every move, a neural engine: with a free account, the analysis is ready seconds after the match ends, each mistake named, explained in plain English, then turned into exercises.
- The bot, no sign-up
- Ranked matches with Elo
- Free, nothing to install
Matches that count.
Pick a length, find an opponent, play against the clock. Your Elo moves after every ranked match, and your game is analyzed while you play.
Ranked, with an Elo rating
1, 3, 5 or 7 points, with the cube from 3 points, 2 minutes per point plus 12 seconds per move. Your Elo, weighted by match length, appears after five placement matches.
points
per point, + 12 s per move
placement matches, then your Elo
Analyzed while you play
With a free account, the engine works during the match. Seconds after the final roll, the full analysis is waiting in the Engine.
- 1You play
- 2The engine analyzes, move by move
- The analysis is ready
Friends, live
Add them by username, see who's online, challenge them with your own settings: 1 to 13 points, cube or not, 1 to 3 minutes per point. Every move shows up on both boards in real time.
Practice mode
Against the bot or a friend: every move graded live, the best move drawn on the board, both players' PR updated as you play.
The bot, any time
Three levels, no sign-up: warm up, try an opening, or play a full match against it at full strength.
Analysis at full strength. Explained in plain words.
The Megammon Engine's neural network evaluates every decision at 3-ply and settles close calls with rollouts. Then Megammon does what an equity column can't: it tells you why, and makes you practice.
The why, not just the how much
16 mistake causes, from “Back checker to escape” to “Wrong take”. Your mistakes get their likely cause, inferred from facts measured on the position: hitting rolls, primes, anchors, timing, race.
Your weakness profile
Your top 3 weaknesses across all your matches, ranked by equity lost, not by frequency. By game phase and by cube decision.
Illustration
Training on your own mistakes
Replay the positions you got wrong. Spaced repetition (FSRS) brings each one back right before you'd forget it.
Every number, decoded
PR, luck rating, match winning chances, a clickable win-probability chart and a tooltip on every stat. Need more precision? “Go deeper” reruns any decision as a 3T rollout.
Illustration
Your matches, your positions
Import .mat, .sgf or text files from your usual software, or paste an XGID and play your move with the mouse.
From the last roll to the right instinct
Five steps, chained automatically.
- 1
Play or import
Play online, with friends or against the bot: with an account, the match is analyzed as you play. Or drop in your .mat, .sgf or text files.
- 2
Neural analysis
The Megammon Engine's neural network evaluates every move (3-ply search) and runs rollouts on close decisions.
- 3
Diagnosis
Your mistakes get their likely cause (escape, too passive, wrong take…), based on facts measured on the position.
- 4
AI explanation
An AI coach explains the mistake in plain English. Every number it cites is checked against the analysis.
- 5
Training
Your mistakes become exercises, scheduled with spaced repetition until the right move becomes second nature.
Expert-level analysis. Without the learning curve.
And you still get every number: equities, probabilities, rollouts, MWC.
An AI that explains. A validator that checks.
A wrong explanation is worse than none. So the AI never gets the last word: every number it cites is checked against the engine's analysis before you see it.
- 1. The facts, measured by the engineWin and gammon probabilities, hitting rolls, back checkers, race: everything is computed before the AI writes a single word.
- 2. The writing, by ClaudeThe model receives only these facts and explains them in 2 to 4 sentences, in plain English, like a coach.
- 3. The check, by a validatorEvery number and every move it cites must appear in the facts. If not, the answer is rejected and rewritten once; failing that, an automatic explanation takes its place.
You played 13/10(2) 10/4, a move that was too passive when the position called for action: 13/10(2) 6/3 6/3* hit and kept the initiative. Your choice exposes you to 36 opponent rolls that hit, versus 16 with the best move, and your winning chances drop to 37.4% instead of 57.7%. You also give up some of your gammon wins (10.9% versus 22.8%), which explains an equity loss of 0.564, a blunder.
Hover over a number to see where it comes from. Written in French by Claude Sonnet, verified by Megammon's validator.
agreement with human labeling on the main cause, across 20 decisions never seen during development
mistake causes recognized, from checker play to the cube
distinct rolls replayed one by one to count every opponent move that hits
numbers shown without being found in the analysis
Learn backgammon, then try it at the board
Short, practical guides to the opening moves, the strategy and the numbers behind each decision. Read one, then test it against the bot.
What you want to know
Is Megammon free?+
Yes: playing online, with friends or against the bot, the analysis and the training are free. A free account keeps your matches, your Elo and your exercises; the bot doesn't even need one.
How does ranked play work?+
Choose a length (1, 3, 5 or 7 points) and start a search: the match begins as soon as an opponent is found. Ranked matches use the cube (except at 1 point) and a clock of 2 minutes per point plus 12 seconds per move. Your Elo, weighted by match length, appears after five placement matches, during which it moves four times faster.
Do I need to install anything?+
No. Megammon runs in your browser. Sign up with an email address or with Google, and your matches are analyzed as you play.
What is Practice mode?+
An option for games against the bot or a friend: every move of both players is graded live at 3-ply, the best move is drawn on the board, and each player's PR updates as you play. The fastest way to learn while playing.
Which engine analyzes my games?+
The Megammon Engine. Its neural network evaluates every decision at 3-ply, close decisions are settled by rollout, and the “Go deeper” button reruns a decision as a 3T rollout.
Which files can I import?+
Matches in .mat format (exported by JellyFish, eXtreme Gammon, Backgammon Galaxy…), .sgf (GNU Backgammon) and text. For a single position, an XGID or a GNUbg Position ID with its Match ID.
Who can see my games?+
Only you: your matches, analyses and exercises are tied to your account. To write an explanation, the AI receives only the numerical facts of the decision being explained (position, moves, probabilities), without the players' names; these facts are sent to Anthropic. If the AI is unavailable, an automatic explanation, computed without AI, is shown instead.
Is the diagnosis reliable?+
Labels are assigned by deterministic rules and measured against human labeling: 70% agreement on the main cause, on 20 decisions held out during development. It's a diagnosis, not a verdict: rate it with a 👍 or a 👎 to help us improve it.
How does training work?+
Your errors and blunders become exercises: you replay the position, the engine grades your move. The FSRS algorithm decides when each exercise comes back: soon if you miss it, later and later as you get it right.
Your next match starts now.
Play online, challenge a friend or warm up against the bot. The analysis follows, automatically.