2048

How to Play

The board is a 4x4 grid. Each move — arrow keys, WASD, or a swipe on touch — slides every tile as far as it can go in that direction. When two tiles holding the same number end up against each other, they merge into one tile worth their combined value, and that value is added to your score. Reach the 2048 tile to win, then choose Keep Playing to chase a higher score, or press New Game to start over. The run ends when the board is full and no two neighboring tiles share a value.

Three rules do most of the work and are worth stating precisely, because strategy follows from them:

• A tile can only merge once per move. A row of 2-2-2-2 becomes 4-4, never a single 8. Merges resolve from the edge you are moving toward, so 4-2-2 pushed left becomes 4-4, and 2-2-4 pushed left also becomes 4-4.
• A new tile appears only if the move changed something. Pressing a direction that slides nothing and merges nothing is a free no-op, not a wasted turn. This is the single most useful rule in the game.
• New tiles are a 2 with 90% probability and a 4 with 10%, placed uniformly at random in one of the empty cells. Two tiles are placed at the start of a game.

Undo — the button or the Z key — rewinds exactly one move, restoring both the board and the score. It does not stack, so once you undo, the next undo is unavailable until you have made another move.

Why the Corner Strategy Works

Almost every piece of 2048 advice reduces to one instruction: pick a corner, put your largest tile there, and never let it leave. The reason is structural rather than stylistic.

Your largest tile is the one thing on the board that cannot be merged away, because nothing else matches it until you build a second one. It is dead weight — a cell permanently occupied by something inert. If that dead weight sits in the middle of the grid, it splits the board: rows and columns that pass through it are blocked, and tiles that need to reach each other cannot. A corner is the only position where a tile blocks the fewest possible paths, since it borders just two cells instead of three or four.

The same argument applies recursively. Your second-largest tile is nearly as inert as the largest, so it belongs adjacent to it along an edge. Your third belongs next to that. Follow the logic all the way down and you arrive at the standard shape: a descending run of values along one edge — 1024, 512, 256, 128 — with the small, active, mergeable tiles kept on the opposite side of the board where they have room to move.

That ordering is not just tidy. It means every large tile has its merge partner directly beside it. When you eventually build a second 256, it sits next to the first, one push away from becoming 512, which is already next to the existing 512. A single well-timed move can cascade through the entire chain. A board where the big tiles are scattered has no such cascades available, and each merge has to be engineered separately.

Banning a Direction

The corner discipline only holds if you stop making the moves that break it, and the practical form of that is to treat one direction as forbidden.

Say you have chosen the bottom-left corner. Your big tiles live along the bottom row. Pressing up lifts every tile off that row — including your anchor. The gap left behind is filled by a new random tile, and now your 1024 is stranded in the middle of the board with a 4 underneath it. There is often no way back. So: up is banned. You play left, down, and right, and you use up only when nothing else moves at all.

This is why the discipline is usually described as avoiding one of the vertical directions rather than a diagonal or a pair. Two of your three permitted directions — down and left in this example — actively press tiles into the anchor and reinforce the structure. Right is the release valve that lets small tiles spread out and find partners, and it is safe as long as your bottom row is full, because a full row cannot slide.

That last condition is the discipline in one sentence: keep your anchor row full. A full bottom row cannot shift when you press left or right, so your big tiles stay exactly where they are while the rest of the board rearranges. The moment a hole opens in that row, sideways moves start dragging your chain out of alignment, and a forced up becomes much more likely.

The Snake Ordering

Once your bottom row is a clean descending chain, the natural next question is where the fifth-largest tile goes. The answer that most strong players converge on is a serpentine, or snake, layout.

Order the cells like a boustrophedon path: left to right along the bottom row, then right to left along the row above it, then left to right along the row above that. Place your values in descending order along that path. The result is that every tile in the sequence is physically adjacent to the next one, including at the row transitions, so the whole board becomes one long merge chain rather than four disconnected rows.

The snake matters most in the endgame. When the board is nearly full, a properly ordered snake can collapse in a chain reaction: the smallest pair merges, which creates a match for the next value up, which creates a match for the one above that, and a single move can clear half the board. An unordered board in the same situation is simply stuck.

Building it is mostly about where you let new tiles land. Small tiles want to live at the tail of the snake, furthest from the anchor. If you keep pushing toward the anchor and use the release direction sparingly, new spawns naturally accumulate at the far end, which is exactly where the sequence wants them.

The Arithmetic of Reaching 2048

It helps to know how much work a 2048 tile actually represents, because it explains why patience beats improvisation.

Every tile on the board traces back to a spawn, and value only enters the game through spawns. If every spawn were a 2, a 2048 tile would be built from 1024 of them. Combining 1024 items in pairs, then pairs of pairs, and so on, takes 1023 merges. Since a new tile appears after essentially every move that changes the board, that is on the order of a thousand moves for a single winning run — and more in practice, because the board is never empty at the end.

The 4s help a little. With 10% of spawns arriving as 4s, the average spawn is worth 2.2, so accumulating 2048 points of raw value takes roughly 930 spawns rather than 1024. Every 4 that spawns is one merge you did not have to perform.

The score follows the same structure. Because a merge adds its result value to your score, and every value from 4 up to 2048 must be produced on the way, the merges that build one 2048 tile contribute about 20,000 points on their own — roughly 2,048 points at each of the ten doubling steps. So if you are at 6,000 points and expecting the 2048 tile shortly, you are not close. You are about a quarter of the way there. Anyone who finishes a winning run has a score above 20,000, usually well above it once the leftover tiles are counted.

How Games Are Actually Lost

Losses are rarely sudden. They are almost always the result of a structural mistake made twenty moves earlier, and they come in a small number of recognizable shapes.

• The panic move. Nothing looks good, so you press the banned direction. Your anchor lifts, a random tile fills the corner, and your chain is broken. This causes more losses than every other mistake combined. If you feel the urge, use one of your other two directions even if it accomplishes very little.
• A trapped small tile. A 2 or a 4 gets stranded between two large, unmergeable tiles in the middle of your chain. It cannot merge with anything and it will not move, so it occupies a cell permanently and cuts your chain in half. Preventing this is most of what good play is.
• Out-of-order chains. You end up with 256, 64, 128 along the edge instead of 256, 128, 64. Now the 64 blocks the 128 from reaching the 256, and reordering requires moves that risk the whole structure.
• Two half-built halves. Building two separate large tiles on opposite sides of the board splits your resources; neither reaches critical mass, and the middle fills with debris. Commit to one chain.
• Filling the board with distinct values. Sixteen tiles that are all different is an instant loss regardless of how big they are. When free cells drop below four, prioritize any merge that opens space over any merge that builds value.

Tips

• Choose your corner before your first move and do not change it mid-game. Relocating an anchor costs more than any single mistake it might fix.
• Keep your anchor row full. As long as it cannot slide, moves along that axis are free and your big tiles stay put.
• Push toward your anchor corner freely — those are the moves that reinforce the structure. It is the opposite direction, the one that lifts tiles off your anchor row, that you should refuse.
• Watch your empty-cell count more closely than your score. Below three or four free cells you are one bad spawn from having no safe move left.
• A move that changes nothing spawns nothing. When you want to see what a direction does without risk, remember that a blocked direction costs you literally nothing.
• Save the undo for a move that genuinely breaks your structure — an accidental banned-direction press — rather than spending it on a merge that merely came out slightly worse than you hoped. It only holds one step.
• Do not chase every merge on offer. Merging two 4s in the middle of the board when your chain needs a different move is how debris ends up in the wrong place.
• Slow down near the end. Early play is nearly automatic and late play is nearly all calculation; most runs are decided in the last thirty moves.

About This Version

2048 is a sliding-tile puzzle with an unusually good ratio of rule simplicity to strategic depth: one action, one merge rule, and a randomness model you can state in a sentence, yet games that reward planning several moves ahead.

This version runs in the browser with no download and no account, on desktop with the arrow keys or WASD and on mobile with swipes. Tiles slide with smooth animation, merges pop with a scale bounce, and the screen shake and sound both scale with the size of the merge — chaining several merges in a row raises the pitch with each one, so a big cascade sounds like what it is. A single-step undo is available from the button or the Z key.

Your current score updates live and your best score is saved locally in your browser, so it persists between visits. Reaching 2048 shows a win screen with the option to keep playing, which is where the real challenge starts: 4096 needs another full run's worth of merges on a board that is already crowded, and every doubling after that costs as much as everything before it combined.