← Back to Ultimate Tic Tac Toe

About Ultimate Tic Tac Toe

Learn more about the game on Wikipedia.

How to Play

Objective: Win three small boards in a row (horizontally, vertically, or diagonally) on the large 3x3 meta-board.

Setup: The game is played on a 3x3 grid of smaller 3x3 Tic Tac Toe boards, creating 81 total cells.

Gameplay:

The AI

It works like DeepMind's AlphaZero. A neural network learned the game by playing itself, with no human games or tips. On its turn, a tree search uses the network to size up moves and plays the best one it found.

It all runs in your browser. Expert thinks for 2 seconds a move, so it's a bit stronger on a faster device, and Very strong looks a few moves ahead. Both first check for a win within their next two turns, so they never miss one. The other levels don't look ahead at all. Strong plays whatever looks best right now, and each level below it picks more at random, down to Random, which plays any legal move. The Analysis page shows what it makes of any position.

Its estimate of who's winning counts draws too: between good players, about half the games are drawn.