FreeCell: the solitaire where almost no deal is a lost cause
FreeCell has a reputation: nearly every deal can be won. We wanted to know whether that holds up on this site's version, so we set a solver loose on 10,000 fresh deals. It found a complete solution for 9,968 of them, which is 99.68%. It did not prove a single deal impossible. The last 32 deals were simply too tangled to finish within the search limit, even on a second, five-times-longer attempt.
That makes FreeCell the game with the highest ceiling of any we tested. All 52 cards are dealt face up from the start, so, unlike Klondike, there is no hidden information for luck to hide behind. When a FreeCell game goes wrong, it is almost always a planning problem, which is good news, because planning can be learned.
You can play it here in your browser for free, on a computer, tablet or phone, with nothing to install and no account needed.
How high the FreeCell ceiling goes
| Result | Value | Deals | How |
|---|---|---|---|
| Solved | 99.68% (95% range 99.55–99.77%) | 10,000 | Solver, 200,000 positions, then 1,000,000 for any left over |
| Proven impossible | 0 deals | 10,000 | Same solver |
| Still undecided | 0.32% (32 deals) | 10,000 | Limit reached twice |
| Published winnable share | 99.998881% (±0.000207%) | about 10,000,000 | Blake & Gent, 2026 |
| Length of the first solution found | about 62 moves on average | 9,913 solved deals | Not the shortest possible |
The published figure comes from Blake & Gent (2026), Journal of Artificial Intelligence Research. It suggests only about one deal in 90,000 is truly impossible, which fits our result of none in 10,000.
Where the ceiling really comes from: the free cells
We wanted to see what makes FreeCell so generous, so we took cells away. Same solver, 3,000 deals for each setting:
| Free cells | Solved | Proven impossible | Undecided | Published winnable |
|---|---|---|---|---|
| 4 (normal) | 99.68%* | 0% | 0.32% | 99.998881% |
| 3 | 96.0% | 0.97% | 3.0% | 99.3583% |
| 2 | 74.0% | 20.2% | 5.8% | 79.544% |
| 1 | 17.4% | 79.6% | 3.1% | 19.348% |
| 0 | 0.2% | 99.7% | 0.07% | 0.2137% |
*10,000 deals with a second pass. Other rows: 3,000 deals, single pass.
The ceiling collapses as the cells disappear. With two cells, one deal in five is provably dead. With one, four in five. Every published figure falls inside the range between our "solved" and "solved plus undecided" columns.
The practical takeaway: each cell you fill during play is a cell you have temporarily lost. Fill three and you are playing a version of the game where 20% of positions are hopeless. Keep them empty whenever you can.
The rules our solver played by
- All 52 cards are dealt face up into eight columns: four of seven cards and four of six.
- Four free cells (top left) each hold one card. Four foundations (top right) build up by suit, Ace to King.
- Columns build down in alternating colors.
- Any card may go into an empty column.
- You can drag a run of several cards at once. The game allows (empty free cells + empty columns + 1) cards per move, which is the same as moving them one by one through the spaces.
Playing toward the ceiling
Plan before the first move
Everything is visible. Take thirty seconds to find all four Aces and all four 2s and count what sits on top of each one. Only 49% of our deals had an Ace ready to play on the very first move, so most games start with some digging.
Aim for an empty column
An empty column is worth more than a free cell, because it also adds one to how many cards you can move at once: on this site the limit is empty free cells + empty columns + 1. Look for the shortest column with the fewest awkward cards and work toward clearing it.
Never fill the last cell without a payoff
Four full cells and no empty column leaves you able to move one card at a time. Only use your last cell if the move that follows wins something concrete.
Don't rush cards home
A card sent to the foundation cannot come back to receive a lower card of the other color. Hold a card back while lower opposite-color cards are still buried.
When stuck, undo rather than restart
Our solver never proved a four-cell deal impossible. If you are stuck, the way out is almost certainly several moves back. Unlimited Undo is there for exactly this.
FreeCell on the ceiling ladder
FreeCell tops every game we measured:
- FreeCell: 99.68% solved, none proven impossible.
- Golf: 92.7% winnable, exact.
- Klondike Turn 1: 76.6–97.9% winnable.
- Baker's Solitaire: 31.8% winnable, exact.
Baker's uses exactly the same deal as FreeCell, which makes the contrast striking: the same 52 cards in the same eight columns, but two rule changes take the ceiling from nearly everything to under a third.
Inside the test, one step at a time
The solitaire-online.us team ran this study in September 2026 to find the ceiling for FreeCell as played on this site.
- Reading the rules from our game code. Eight columns, four free cells, alternating-color building and any card into an empty column, all taken from this page's game files.
- Rebuilding the game. We wrote FreeCell as a program with those rules. The move-several-cards rule does not change which deals can be won, because the same cards could always be moved one at a time.
- Dealing the cards. Seeds 1 to 10,000 for the main test and 1 to 3,000 for each reduced-cell test. These are our own numbers, not the old Windows deal numbers.
- Choosing the player. A solver that searches for a complete solution. FreeCell has no hidden cards, so the solver knows exactly what you know.
- Running it. Each deal got up to 200,000 positions. The 87 deals still undecided after that were run again with 1,000,000; 32 remained.
- Checking against published research and a second program. Every published figure from Blake & Gent (2026) lies within our ranges. We also re-checked some "impossible" verdicts with a completely separate brute-force program: 5 deals with one cell and 2 deals with three cells. Both programs agreed every time.
- Working out the margin of error. A 95% range means the true value would fall inside it in 19 of 20 repeats. For four cells, deals solved: 99.55–99.77%.
- What the numbers cannot tell you. "Solved" means a solution exists, not that a person will find it. The site also lets a card come back off a foundation, which our solver never used; that can only make our figures slightly low.
The code behind the ceiling
We wrote the study in Python 3.11, standard library only. FreeCell and Baker's Solitaire share one module, freecell.py, and runner.py spreads deals across 19 worker processes on a 20-core machine. A second script, retry.py, re-ran the undecided deals with the larger limit, and verify_freecell_bruteforce.py is the independent cross-check.
At each step the solver picks the position that looks closest to a win, ranked by how many cards are not yet home and how many sit on a lower card. It stores each position in a standard form so it never explores the same one twice, and it makes only foundation moves that are provably safe. The main 10,000-deal run took about 18 minutes; the reduced-cell runs took from about 1 minute (no cells) to about 25 minutes (two cells). Ranges use the Wilson score interval.
We had help from an AI assistant, Claude by Anthropic, in writing and reviewing the code and in drafting this page. The AI did not generate any of the numbers. They all come from running the programs, and we compared them with the published figures before publishing.
Because each shuffle comes from a numbered seed, anyone with the same code gets exactly the same deals and the same verdicts.
By Admin, solitaire-online.us. Study date: September 18, 2026.
Ceiling questions from players
Has anyone found an impossible FreeCell deal?
Almost. In our test, none of 10,000 deals was proven impossible. The published estimate is 99.998881% winnable, so a truly impossible deal is extremely rare.
Why do I lose, then?
Most often, the cells ran out before the columns were sorted. With fewer usable cells, the share of winnable positions drops fast, as the table above shows.
How long is a solver solution?
The first solution our solver found averaged about 62 moves. People typically take more, and that is fine.
Which has the higher ceiling, FreeCell or Klondike?
Easier to win, harder to play carelessly. Compare the ceiling here with Klondike Turn 1, where about one deal in ten cannot be won.
What's a tougher game with the same layout?
Baker's Solitaire uses the same deal but builds by suit and only lets Kings fill empty columns. On this site, fewer than a third of its deals can be won. See every game on the games page.