SCANNER TECHSJASSI SINGH
← Blog

Is 2048 Luck Based? I Played 23,000 Games

Is 2048 luck based? Mostly not. I ran 23,000 games of 2048 through players of different strength, and the gap between them was enormous: making random moves never produced a 1024 tile in 10,000 games, while a player that looked two moves ahead made 2048 in 94% of its games. The new tiles are random, so luck settles some games, but how far ahead you think settles almost all of them.

I make a browser version of 2048, and its rules run on their own outside the browser, so I could play thousands of games instead of guessing, as I did for Minesweeper and luck and Klondike Solitaire.

The numbers

Four players, each on the game’s own rules and its own random tiles, each playing until no move was left:

  • Random: any move that does something, picked at random.
  • Corner: the rule people are usually taught. Down if it can, else left, else right, else up.
  • One move ahead: tries each move and keeps the board that looks best.
  • Two moves ahead: tries each move, then every square the new tile could land in, as a 2 or a 4, then every reply, and picks the move with the best average. On a crowded board it looks three moves ahead.
Player 512 1024 2048 4096 8192
Random 0.02% 0% 0% 0% 0%
Corner 4.6% 0% 0% 0% 0%
One move ahead 30.7% 1.6% 0% 0% 0%
Two moves ahead 99.8% 98.7% 94.3% 66.4% 10.9%

The percentages are the share of games that made each tile at least once. The random and corner players played 10,000 games each, one move ahead 2,000 and two moves ahead 1,000.

The random player usually stalled at 64 or 128. Its median score was 1,048. Two games in 10,000 made a 512.

The corner rule did better, with a median score of 2,416: half its games made a 256, but none of 10,000 made a 1024. Following the rule blindly keeps the big tile in the corner, but it can’t stop a 2 landing in the one square that wrecks the row.

Looking one move ahead helped again, to a 512 in about a third of games. The jump came from looking two moves ahead and weighing where the new tile might land. That is the difference between a player that never made 2048 in 2,000 games and one that made it 943 times out of 1,000.

How I measured it

Every game was played by the game’s own code: the same sliding rules and the same random new tiles as the page, starting from game number 1 for every player. The players that look ahead think with a fast copy of the sliding rules, which I checked against the game’s code on 300 games before using it.

To judge a board, they use the scoring most strong 2048 programs share, from Robert Xiao’s 2014 AI: points for empty squares and for tiles that can join, and penalties for rows and columns that don’t rise steadily one way and for big tiles spread apart. The two-moves-ahead player took about 77 minutes for its 1,000 games on my Mac.

What this doesn’t measure is how often people win. A person sees patterns these players don’t, and makes mistakes they don’t. It shows how much of the game is decided by thinking ahead, against how much is decided by the tiles.

What perfect play looks like

Two other results show where the ceiling is.

Xiao’s AI searches four to eight moves ahead instead of two. In his write-up, it made 2048, 4096 and 8192 in all 100 of its games, 16384 in 94 and 32768 in 36. Its lowest score was 124,024.

In 2023, Alexey Slizkov proved the first hard numbers. With the best possible play, the chance of reaching 2048 is at least 99.969%, so luck can beat a perfect player in at most about 3 games in 10,000. He also proved that a player can always reach 256, whatever tiles appear and wherever they land. His results suggest it stops well short of 2048: if the new tiles were placed on purpose to stop you, rather than at random, nobody could make 2048.

Where the luck is

The new tile is the only random thing in 2048. In the original game’s code it goes in an empty square picked at random, and it is a 2 nine times in ten and a 4 otherwise. Mine works the same way. It doesn’t look at your board, so the game isn’t rigged against you. It just doesn’t care.

That randomness matters more the higher you aim:

  • Up to 1024, luck hardly counts. The two-moves-ahead player made 1024 in 987 games of 1,000.
  • At 2048, luck starts to show. The same player missed it in about 1 game in 18, and Slizkov’s bound says even perfect play misses now and then.
  • At 8192 and beyond, luck is a big part of it. My player made 8192 in 11% of games and never 16384. Xiao’s much deeper search still missed 32768 in nearly two games of three.

How many moves it takes

About 1,000. Each move adds one tile worth 2.2 on average, and joining tiles never changes their total, so a board that holds a 2048 tile has had at least 2048 points of new tiles added to it. That takes at least about 930 moves, fewer only when a lot of 4s arrive. In my test, games that made 2048 took between 926 and 1,280 moves, with a median of 983.

The score at that moment is usually close to 20,000. Making a 2048 tile out of 2s takes ten rounds of joins, and each round adds 2048 to the score, a little less for every tile that arrived as a 4.

How to need less luck

What separated the players above:

  • Keep the biggest tile in a corner, and play mostly the two directions that keep it there. Use a third when you must, and avoid the fourth.
  • Keep the row along that wall full, with the numbers falling away from the corner, so joins run toward it one after another.
  • Look at where the new tile can land. This is what the corner rule misses and the look-ahead player does. Before a move, check which squares it will leave empty, and keep them away from the row you are building.
  • Keep empty squares. Every strong player, Xiao’s included, scores a board largely on how many squares are free. A full board is one bad tile from the end.

Play it

My 2048 is free in the browser on a phone or a computer, with no download. It’s the classic game: a four by four board, the original rule for new tiles, and Keep going after 2048 for 4096 and beyond. Swipe on a phone, or use the arrow keys or W, A, S and D. The game in progress is saved in your browser, so you can close the page and come back to it, and your best score is kept on your device.

The honest limits, as of November 2026:

  • There’s no undo during a game. At the end you may be offered an ad that takes back the last three moves, at most once per game.
  • It’s the classic four by four board only, with no other sizes and no daily puzzle yet.
  • Scores are kept on your device only. There’s no account and no online leaderboard.

Questions people ask

Is 2048 luck or skill?

Mostly skill. In my test of 23,000 games on the standard rules, a player making random moves never made a 1024 tile, while a player that looked two moves ahead made 2048 in 94% of its games. The new tiles are random, so luck decides a few games, but how far ahead you think decides almost all of them.

Is 2048 always winnable?

Not quite, but almost. A 2023 paper by Alexey Slizkov proved that the best play reaches 2048 with a probability of at least 99.969%. It also proved you can always reach 256 whatever tiles appear, and its results suggest that if the tiles were placed to stop you on purpose, nobody could make 2048.

Is 2048 rigged?

No. In the original game, each new tile goes in an empty square picked at random, and it is a 2 nine times in ten and a 4 otherwise. It doesn't look at your board or try to block you. Bad runs happen because the board fills up, not because the game is against you.

How many moves does it take to get 2048?

About 1,000. One new tile appears per move and is worth 2.2 on average, and the tiles on the board add up to at least 2048 when you make it, so it can't take much fewer than 930 moves. In my test, games that made 2048 took between 926 and 1,280 moves, with a median of 983.

What is a good 2048 score?

The score at the moment you make 2048 is usually around 20,000, because building a 2048 tile out of 2s adds 2048 to your score ten times over. In my test, a random player's median score was 1,048, a player always pushing toward one corner got 2,416, and a player looking two moves ahead got 62,872 by playing on past 2048.

What is the best strategy for 2048?

Keep your biggest tile in a corner, play mostly two directions, keep the row along that wall full with the numbers falling away from the corner, and before each move look at where the new tile could land. The corner rule alone is not enough. In my test, a player that only followed it never made a 1024.

The game

2048BrowserThe sliding number puzzle: join matching tiles and make 2048. Your game is saved as you play.