This 2048 AI (artificial intelligence that can solve 2048 positions one move at a time) plays the game for you or suggests a single move. Press Watch AI play to let it run, change its speed, and pause whenever you like. Press Best move to get one suggestion with a short reason. You can also bring over the board from your current 2048 game. Everything runs in your browser; nothing is sent to a server.
How the 2048 solver works
The bot uses a search algorithm called expectimax. For each of the four moves, it looks at what the board would become, then at where the next random tile could appear (a 2 nine times out of ten, a 4 otherwise), then at its own best reply, a few steps ahead. It averages the random outcomes and picks the move with the best expected result.
To judge a position at the end of that search, it scores four things: how many cells are empty, whether rows and columns rise or fall in order, whether neighbouring tiles are close in value, and whether the biggest tile is in a corner. These are the same habits described in how to win 2048, written as arithmetic.
How strong is it?
It is tuned to answer in a fraction of a second on a normal laptop, so it searches only two or three moves ahead. In 13 test games it reached the 2048 tile every time and 4096 in 6 of them; none reached 8192. The strongest published 2048 programs search far deeper and use large precomputed tables; they reach 16,384 in 99.9% of games. See the 2048 high score page for their numbers and for the highest score that is possible at all.
Using Best move without spoiling the game
Try to decide your move first, then press Best move and compare. When the AI disagrees, read its reason. Most disagreements come from the same few mistakes: pulling the big tile out of its corner, leaving a gap in the edge row, or ignoring a stray 2.
Questions
Is using a 2048 AI cheating?
Only if you are competing with someone. For learning, watching a solver and comparing its moves with yours is one of the quickest ways to improve.
Can the AI always win 2048?
No. The new tiles are random, and some positions cannot be saved. Even the strongest programs occasionally lose before 2048.
Why does the AI sometimes pause on a move?
When the board is nearly full, it looks further ahead to avoid a losing move, which takes a little longer.