Learn backgammon by playing. Games against an AI that scales from someone who has just learned the rules to world class, with a coach that grades every move, explains what happened, and shows you what it would have played.
- An opponent at your level. Nine AI levels from Pebble to Phoenix, all played by GNU Backgammon's neural network. Win and your next game is a level up; lose and it steps down. Play Light, Dark, or both sides in study mode, where you can also set the dice by hand.
- A coach that watches every move. Win bar, expected result, and a grade for each move, from Best to Blunder, with a plain-language reason: a hit you missed, a shot you left, a point you could have made, the race, the gammon risk. It runs right in your browser. Set Coach explains for to match your experience: beginners hear games won and lost out of 100, stronger players get the equity. Point at Try it instead to see the coach's move on the board, click to play it, and Back to my move to return. Turn on Find better moves yourself to get another try after a mistake before the coach shows its move.
- The doubling cube. The coach grades your doubles, takes and passes, and tells you when you should have doubled. Or switch the cube off and just play.
- Luck kept apart from skill. The coach tells you when a roll was very lucky or unlucky, so you can tell a bad move from bad dice.
- Accessible: play by keyboard or by ear. Tab to the board and play with the arrow keys and Enter; every point you move to is described. Screen readers hear every roll, move and coach comment, or turn on Speak announcements to have them read aloud.
- Hints when you want them. Ask for the coach's best moves for your roll, or press Their idea to see what your opponent hopes to roll next and what it could hit.
- See the board like a stronger player. Overlays for pip counts, shots (how many rolls hit each blot), points to make, where a checker can land, the last play, the best play and point numbers.
- Take back freely. Undo your move and play the same roll again, or step back through the whole game.
- Review your games. A game graph with mistakes and lucky rolls marked. Click to jump to any move. After each game, Key moments picks the few decisions most worth a look. Save and load games; your current game and your score are saved automatically.
A one-minute guide to the rules, a short guide to strategy and the keyboard shortcuts are in the game itself.
Every level is GNU Backgammon's neural network, rebuilt to run in your browser and checked against GNU Backgammon itself: on 300 test positions they pick the same move in all but one, and make the same cube decision in every case tried.
At the top, it plays as well as the best human players. Dragon plays like GNU Backgammon's "World Class" setting, which its makers describe as close to the best players in the world, and Phoenix, looking at more moves, a little better still. Of the backgammon programs, only eXtreme Gammon is generally rated stronger. Phoenix thinks for at most 10 seconds a move, so on slow devices it reads less.
Below Dragon the levels are held back on purpose: they look less far ahead, choose among their moves more loosely, and the lowest ones fall back on a beginner's habits, hitting whenever they can or running for home. River plays like GNU Backgammon's "Expert" setting, a strong club player; Pebble has just learned the rules. In self-play each of the lower levels beat the one below it about 7 games out of 10. Dice are dice, so the weaker side still wins about 3 in 10.
© 2026 Frank Force. Free and open source under the GPL-3.0 license. The AI and the coach use the neural network of GNU Backgammon, and its evaluation code is ported from GNU Backgammon's source (notices).