FreeCell

How to play FreeCell

All 52 cards are dealt face up into eight columns — seven cards in the first four, six in the rest. Nothing is hidden and there is no stock to draw from, so there is no luck after the deal. Everything you need to know is on the table from the first second; the only question is whether you can see it.

You are moving cards to the four foundations in the top right, which build up by suit from the ace. Get all four to the king and you have won. In the tableau below, cards stack down in alternating colours — a red six onto a black seven. The four free cells in the top left each hold exactly one card, any card, as a parking space.

The rule everyone gets wrong

How many cards you can move at once. Formally you may only ever move one card at a time; a "supermove" of several cards is shorthand for shuttling them through the free cells and back. So the number you can shift is:

(free cells + 1) × 2^{(empty columns)}

It multiplies. Every empty column doubles the allowance, because a whole sub-sequence can be parked there and the free cells reused around it. Four free cells and two empty columns is twenty cards, not seven. This is the single most botched rule in FreeCell software, and it survives because the two formulas agree at small numbers — with one free cell and one empty column both say 4.

And the half that gets dropped even by implementations that get the doubling right: if you are moving into an empty column, that column does not count, because it cannot be the destination and the staging area at the same time. The allowance halves. Our engine does not just return the number — it constructs the actual sequence of single-card moves, and the test suite replays every constructed sequence through the ordinary one-card rules.

Why a deal becomes impossible, and what we do about it

Nearly every FreeCell deal is winnable. Of the original Microsoft 32,000, exactly one is not: #11982. Of the first million, eight are. So if your game is lost, it is overwhelmingly likely that you lost it — and the usual culprit is a card that went home too early.

A card is safe to send to its foundation only when nothing left in play could still need it as a base. A card of rank R can only ever be built on by the two opposite-colour cards of rank R−1, so it is safe once both opposite-colour foundations have reached R−1. Send a black seven home while a red six is still buried and that red six may have nowhere to go for the rest of the game.

Most implementations autoplay greedily — anything that will go, goes — and will do this to you without a word. Our test suite does not argue this point, it exhibits it: on deal #8, sixty-six moves into that deal's own solution, greedy autoplay sends home one card, the seven of clubs, and the deal becomes provably unwinnable — 976 reachable positions exhausted with no win among them. Safe autoplay, from the byte-identical position, declines that one card and leaves the deal winnable in 41 moves. One card.

So this board watches. After every move it re-solves the position in the background. If the deal is still winnable it says so. If it has become unwinnable it tells you immediately, names the move that did it, and offers to take it back — and it will only ever say that when it has proved it, by exhausting every position reachable from where you are. That is the honest version of a feature, and the honest version has a limit: on a small number of positions the search runs out of budget and resolves nothing, and then the panel says "could not decide" rather than guessing. It never guesses in either direction.

The deal numbers are real

Windows FreeCell generated each of its 32,000 deals from the deal number using a published algorithm: a 32-bit linear congruential generator, seed = seed × 214013 + 2531011, dealing from a 52-card array by swapping with the last undealt card. Windows XP extended the range to a million, keeping the first 32,000 identical. The numbering became the lingua franca of the game — it is how players compare notes about a hard deal, and it is why "11982" means something.

We use that generator, and we check it. The suite asserts our output against the published MSVC rand() sequence for seed 1 — 41, 18467, 6334, 26500, … — and then asserts deals #1 and #617 card for card, in all 52 positions each, against their published layouts. #617 is not an arbitrary choice: Michael Keller's FreeCell FAQ names it as the most asked-about deal after 11982.

In fairness to the competition: several sites already offer a winnable deals only filter, and solitaired.com ships a solvability bot that will tell you whether the shuffle you were dealt can be won. They are good sites and we are not claiming to have invented that. Two things are genuinely different here. First, the check runs continuously on your position, not once on the opening deal — which matters far more, because the deal was almost certainly winnable and the question is whether it still is. Second, the deal numbers are Microsoft's: cardgames.io, one of the biggest free FreeCell sites, lets you type a game number from 1 to 50000 that comes from its own private shuffle and therefore means nothing to anyone else.

Strategy that actually helps

Empty a column before you spend a free cell. An empty column is worth more than a free cell and the formula says exactly how much more: a free cell adds one to what you can move, an empty column doubles it. Two free cells and no empty column moves 3; one free cell and one empty column moves 4.

Dig for the aces and low cards first. Nothing else can start. A deep ace is the whole game; a deep king is almost free, because kings only ever want an empty column.

Do not send everything home. The tableau needs high cards to build on. A five left out is a base for a four; a five sent home is one less place to put things. Send it when it is safe, not when it is possible — and the board's autoplay already follows that rule, so if a card has not gone home by itself there is usually a reason.

Watch for a suit buried under its own lower cards. If the 2 of spades sits beneath the 5 of spades in the same column, that column cannot resolve itself — the 5 cannot go home until the 2 does, and the 2 cannot come out until the 5 moves somewhere else. Those are the columns to break up early, while you still have the room.

Plan the whole move before the first card. Because supermoves are really shuttles through the free cells, a move that leaves your cells full has cost you more than it looks. Count the allowance before you commit, not after.

Questions people ask

How many cards can you move at once in FreeCell?

(free cells + 1) × 2^{(empty columns)}. It is multiplicative, not additive. Two empty free cells and no empty columns moves 3 cards; free a column as well and it is 6, not 4. Four cells and two empty columns is 20. If you are moving into an empty column, that column cannot also be the staging area, so it does not count and the allowance halves.

Is every FreeCell deal winnable?

Almost. Exactly one of the original Microsoft 32,000 is not — deal #11982 — and eight of the first million are not: 11982, 146692, 186216, 455889, 495505, 512118, 517776 and 781948. Our solver proves every one of those eight unsolvable by exhaustive search and solves the rest, so the useful question is not whether the deal was winnable but whether it still is.

What is FreeCell deal 11982?

The famous unsolvable one. From August 1994 to April 1995 Dave Ring organised more than a hundred volunteers on Usenet to solve all 32,000 Windows deals, a hundred each. Every deal fell except 11982. The start screen here offers it deliberately — and the board tells you it cannot be won, because our solver exhausts all 61,643 positions reachable from it without finding a win.

Can you move a card back off the foundation?

Not here, and not under the standard rules — Keller's FreeCell FAQ is explicit that cards played home stay home. Some programs allow it as a variant, called worrying back. We follow the standard rule because every published solvability figure for these deals was computed under it, and because unlimited undo fixes the same mistake without changing the game. It is worth very little in any case: only 11 deals in the first 32,000 become winnable with it, and 11982 is not one of them.

Does this site use the same deal numbers as Microsoft FreeCell?

Yes, and it is asserted rather than claimed. The suite checks the generator against the published MSVC rand() output for seed 1, then checks deals #1 and #617 against their published layouts in all 52 positions each. Deal #617 here is deal #617 in Windows FreeCell.

Why did my FreeCell game become impossible?

Usually a card went to a foundation too early. A card is safe to send home only when both foundations of the opposite colour have reached at least one rank below it, so nothing left in play can need it as a base. Most implementations autoplay greedily and do this to you silently. This one uses the safe rule, and re-checks the whole position after every move so that when a deal does die you find out while undo can still fix it.

Is this FreeCell free?

Yes. No signup, no ads, no video between games, no daily limit, no premium tier, no paid hints, and no ad-free subscription — because there are no ads to remove.

The honest small print

The solver is a best-first search with a transposition table and safe autoplay, not a competition solver. It has three possible answers and it distinguishes them: won (it found a line, and the line is replayed through the ordinary rules before it is trusted), proved dead (it exhausted every reachable position and found no win), and could not decide (it hit its budget). The third is not a soft version of the second and this page never presents it as one. The proof is also a proof about the standard move model — single-card moves, and no card taken back off a foundation. Your best times and win counts are stored unverified, because a browser can lie about its own score and we would rather label that than pretend otherwise. Deals 1–32000 are the original Microsoft set; numbers above that use the same published generator, extended the way Windows XP extended it.