About 82% of draw-3 Klondike deals can be won, if you could see every card. That figure comes from a 2026 paper in the Journal of Artificial Intelligence Research, and it's solid. The catch is the "if." Nobody plays that way. In the solitaire on your phone, most of the cards start face down, and that changes the answer.
Is solitaire a game of skill or luck? Home players swear it's luck. Cash-game players are told it's skill. Both sides quote numbers, and most of those numbers get misread. That includes the famous 36.6% "expert" win rate, which came from a player who also knew every card.
We make a cash solitaire game, so we wanted a straight answer too. We gave three computer players the same 20,000 deals, under draw-1 and draw-3 rules, and measured what better decisions add. Below: what the famous numbers measure, where the luck hides, our test, FreeCell versus Klondike, cash matches, the law, and the strongest case for luck.
- Both, not equally, and the rules set the mix. Draw-3 Klondike carries far more luck than draw-1, and FreeCell sits near the skill end.
- The famous 82% assumes you can see every card. It's the winnable share of draw-3 deals when you already know what's face down (Blake and Gent, JAIR 2026).
- Decisions still move win rates a lot. With cards hidden, random moves won 7.1%, a simple rule set 13.0% and the best planners about 37% (Bjarnason et al., 2009).
- Our test agrees. On the same 20,000 deals, our strategy-rule bot won 51.2% under draw-1 rules and 16.3% under draw-3 (Atay Games bot test, 2026).
- A same-deal cash match equalizes the deal, not the luck. And we found no court ruling on solitaire itself.
So, is solitaire a game of skill or luck?
Both, in a ratio the rules set. The luck lives in the cards you can't see, and your decisions decide how many deals you convert. The gap is measurable. In 2009, a study of hidden-card Klondike found that random moves won 7.1% of games, while the best computer planner won about 37% (Bjarnason, Fern and Tadepalli, ICAPS 2009).
Each side of the kitchen-table argument holds a real fact. One side says most deals are winnable, so it's skill. The other says you can't win a bad deal, so it's luck. Both facts are true. Both get misread, because they describe different games: one where you know every card, and the one you actually play.
Not all evidence is equal, so we ranked it, just as we did when we asked the same question for gin rummy. Most pages that answer this question lean on the bottom row of this table.
| Evidence | What it shows | Strength | Caveat |
|---|---|---|---|
| Controlled bot tests (Bjarnason 2009; Atay 2026) | Better decisions win far more often on the same kind of deals | Strong | Bots, not people |
| Perfect-information solver (Blake and Gent 2026) | How many deals are winnable if you know every card | Strong, for its game | Not the game you play |
| Planner vs expert on "thoughtful" solitaire (Yan et al. 2004) | A computer planner won about twice as often as a top expert when both knew every card | Moderate | Every card known; one expert |
| Platform win rates (Solitaired) | What real players win, by variant | Moderate | How unfinished games count isn't stated |
| Expert opinion and "solitaire is skill" marketing | Confident claims, either way | Weak | No data behind them |
The short verdict. Solitaire is both, in a ratio the rules set. With every card known, 81.9% of draw-3 Klondike deals can be won (Blake and Gent, JAIR 2026). With cards hidden, as in real play, the best published AI wins about 37% (Bjarnason et al., 2009). FreeCell, dealt face up, is about 99.999% winnable.
What does "82% of games are winnable" actually mean?
It's the share of draw-3 deals a perfect player could win if they knew every card in advance. Researchers call that version "thoughtful" solitaire. In 2026, Blake and Gent put it at 81.945% ± 0.084% in the Journal of Artificial Intelligence Research.
Two quick definitions. The stock is the face-down pile you turn cards from. Under draw-3 rules, you turn it three cards at a time and can play only the top one. Under draw-1, you turn one card at a time, so every stock card is within reach.
The abstract is precise about the condition: "we report the winnability of Klondike as 81.945% ± 0.084% (in the 'thoughtful' variant where the player knows the rank and suit of all cards)." Face-down cards still can't be moved until they're turned, which, the authors note, "means that thoughtful Klondike is subtly different from a variant in which all cards start face-up."
Their rules are generous, too. The stock is drawn in groups of three, "redeals are allowed without limit," and cards may be "worried back" from the foundations to the tableau. Under draw-1, the figure is 90.480% ± 0.116%. Here's what each famous number measures.
| Figure | Game and rules | Who plays | What it tells you | Source |
|---|---|---|---|---|
| 90.5% | Klondike draw-1, every card known | Perfect solver | Upper bound for draw-1 | Blake and Gent 2026 |
| 81.9% | Klondike draw-3, every card known | Perfect solver | Upper bound for draw-3 | Blake and Gent 2026 |
| 70.2% | Draw-3, every card known | Rollout planner, 200 games | Strong play, but not perfect | Yan et al. 2004 |
| 36.6% | Draw-3, every card known | One expert, 2,000 games, about 20 minutes each | Even an expert who knows every card loses often | Yan et al. 2004 |
| About 37% | Draw-3, cards hidden (the real game) | Best Monte Carlo planner | A floor on what's winnable in real play | Bjarnason et al. 2009 |
| 33.0% / 11.1% | Draw-1 / draw-3, real play | Players on one large site | What people actually record | Solitaired (2024 data) |
What about the famous 36.6% "expert" win rate?
It comes from the same every-card-known game. In a 2004 paper co-written by Stanford's Persi Diaconis, the authors wrote that "our esteemed expert averages about 20 minutes for each game of thoughtful solitaire. He carefully played and recorded 2,000 games, achieving a win rate of 36.6%" (Yan, Diaconis, Rusmevichientong and Van Roy, NIPS 2004). He knew every card. It isn't a normal Klondike win rate, though several pages quote it as one.
That paper opens with a line Blake and Gent echo in their abstract: "It is one of the embarrassments of applied mathematics that we cannot determine the odds of winning the common game of solitaire." For the hidden-card game, it's still true. The share of real draw-3 deals a perfect player could win sits somewhere between about 37%, which the best published AI already wins, and 82%, the winnable share with every card known.
What 82% measures. The popular "82% winnable" figure describes thoughtful Klondike, where the player knows every face-down card. Blake and Gent report 81.945% for draw-3 and 90.480% for draw-1 in the Journal of Artificial Intelligence Research (2026). In a 2004 study co-written by Stanford's Persi Diaconis, an expert who knew every card still won only 36.6% of 2,000 games.
Where does the luck in Klondike actually come from?
From what you can't see. At the deal, 45 of Klondike's 52 cards are hidden: 21 face down in the tableau and 24 in the stock. Only 7 are visible. And even with every card known, about 18% of draw-3 deals can't be won, according to Blake and Gent's 2026 figures, so some of the luck is baked into the shuffle.
Card counts are arithmetic from the standard Klondike layout. Unwinnable shares are 100% minus the winnable shares in Blake and Gent, Journal of Artificial Intelligence Research, 2026.
That hidden information creates luck in three ways:
- Unwinnable deals. Even a perfect player who knows every card loses about 18% of draw-3 deals (100% minus 81.945%) and 9.5% of draw-1 deals (100% minus 90.480%). No decision fixes those.
- Forced guesses. Sometimes two moves look equally good, and only the hidden cards decide which one works. Bjarnason and colleagues illustrate one such position in their 2009 paper, captioned "A dead end forced by having to guess."
- Draw-3 access. Under draw-3, you can reach only every third stock card on a pass until your plays shift the order. Under draw-1, you see the whole stock on the first pass.
All three come down to information, not the shuffle as some mysterious force. A good player can't see through a face-down card. They can choose which card to uncover next, and when. That's why FreeCell, which deals every card face up, behaves so differently, as you'll see below.
Where the luck lives. Klondike hides 45 of its 52 cards at the deal: 21 face down in the tableau and 24 in the stock. That hidden information is where the luck lives. Even with perfect play and every card known, about 18% of draw-3 deals and 9.5% of draw-1 deals can't be won (Blake and Gent, 2026).
How much do better decisions raise the win rate?
A lot, even with the cards hidden. In 2009, Oregon State researchers Bjarnason, Fern and Tadepalli played one million hidden-card games for each baseline strategy. A random player won 7.135%, and a greedy player following a fixed priority list won 12.992%. Their Monte Carlo planners, which simulate many possible games before choosing each move, won "over 35%."
Their best version, an ensemble planner, won 36.97% ± 1.92% of 4,206 games at 99% confidence. That's where this post's "about 37%" comes from. The authors didn't undersell it: "These results more than double current estimates regarding human level performance."
The same pattern holds when every card is known. In the 2004 paper, a simple heuristic won 13.05% of thoughtful games, and iterated rollouts, which play out candidate moves before choosing, reached 70.20% over 200 games. That's about twice the expert's 36.6%, and he was no novice: "a senior American mathematician (former president of the American Mathematical Society and indeed a famous combinatorialist)."
The draw-3 Klondike ladder: what decisions add, and what seeing every card adds
Sources: Bjarnason, Fern and Tadepalli, ICAPS 2009 (cards hidden; the 37.0% bar is their best ensemble planner, 36.97% ± 1.92% at 99% confidence). Blake and Gent, Journal of Artificial Intelligence Research, 2026 (every card known).
So the ladder runs from 7.1% to about 37% on the same hidden cards, and decisions alone make the difference. The last bar isn't the next rung. It's a different game, one where the player already knows every card.
Better decisions, more wins. Better decisions multiply wins on the same kind of deal. In 2009, Oregon State researchers found a random-move player won 7.1% of hidden-card Klondike games and a simple rule-following player 13.0%, over one million games each. Their best Monte Carlo planner won about 37%, which they said more than doubled estimates of human performance.
What happened when we gave three bots the same 20,000 deals?
On identical deals, decisions changed the result a lot, and so did the draw rule. In September 2026, Atay Games played 20,000 hidden-card Klondike deals with three bots. Under draw-1, a random-move bot won 26.7%, a published rule-follower 42.5% and our strategy-rule planner 51.2%. Under draw-3, that same planner won 16.3%.
How we ran it
- The game. Standard Klondike with hidden cards: seven columns dealt 1 to 7 with the top card face up, a 24-card stock, building down in alternating colors, only Kings into empty columns, and cards allowed back off the foundations. The stock turns 1 card at a time (draw-1) or 3 (draw-3), with unlimited passes, Windows-style. A game counts as won once nothing is face down and the stock and waste pile are both empty, since from there every game can be finished.
- Three players, same information. Each bot sees only what a person sees: face-up cards, the top waste card, the foundations and how many cards are face down. Random picks any legal move at random. Rule-follower is our port of the published greedy priority order in Bjarnason, Fern and Tadepalli (2009), except that it turns the stock one packet at a time, like a person.
- Our planner. Planner is our own rule set, built from common strategy advice. It plays Aces, Twos and other safe foundation cards at once. It reveals face-down cards first, starting with the column holding the most. It moves runs only to reveal a card or free a foundation card, plays the waste card before turning more stock, and puts a King in an empty column only if that reveals a card.
- Same deals. All three bots played the same 20,000 shuffles (seed 20260923, one fixed seed per deal) under both draw rules. Copies of the rule-follower and planner with different tie-break seeds served as equal-skill controls.
- Stopping rule. As in the 2009 paper, a bot can't make a move that recreates a position it has already seen in that game. A game ends when every legal move would repeat, or at 30,000 moves, which counts as a loss. Only 91 of the 200,000 games hit that cap, 89 of them by the random bot.
- Checks. Our rule-follower won 12.6% of draw-3 games (95% CI 12.1% to 13.1%), in line with the 12.992% the paper reports for its greedy player over a million games. Our random bot (2.7% under draw-3) is weaker than the paper's random player (7.135%), which could reach any stock card directly and switched to greedy play once nothing was face down. A self-test confirming all 52 cards are accounted for after every move passed on 200 deals, 10 games each.
- Run. September 23, 2026, standard-library Python 3.14.4, 207 seconds on 14 processor cores. The script itself isn't published.
Win rate on the same 20,000 deals, by bot and draw rule
Atay Games bot test, September 23, 2026 (first-party data). A model with bots, not a measurement of people or of any app. Method in the box above.
What did the results show?
Decisions mattered on identical cards. Under draw-1, moving from random moves to our planner nearly doubled wins on the same deals, from 26.7% to 51.2%. Under draw-3, it multiplied them six times, from 2.7% to 16.3%. The planner also got more cards home: 30.1 of 52 on average under draw-1, against the random bot's 18.3.
| Player (same 20,000 deals) | Draw-1 win rate (95% CI) | Draw-3 win rate (95% CI) | Avg cards on foundations, draw-1 / draw-3 |
|---|---|---|---|
| Random (any legal move) | 26.7% (26.1 to 27.3) | 2.7% (2.5 to 2.9) | 18.3 / 4.6 |
| Rule-follower (published greedy order) | 42.5% (41.8 to 43.2) | 12.6% (12.1 to 13.1) | 26.2 / 10.9 |
| Planner (strategy-advice rules) | 51.2% (50.5 to 51.9) | 16.3% (15.8 to 16.8) | 30.1 / 12.7 |
| Equal-skill copies (different tie-breaks) | Rule-follower 42.5%, Planner 51.2% | Rule-follower 12.9%, Planner 16.4% | Matches the originals, as it should |
Deal by deal, the pattern held. Under draw-1, the planner won 12.4% of deals that the rule-follower lost, while the reverse happened on only 3.8%. Both won 38.7% of deals, and both lost 45.1%.
Which mattered more, the bot or the rules? The draw rule was the bigger dial. Switching the same planner from draw-1 to draw-3 cut its wins by about two-thirds, from 51.2% to 16.3%, on the very same shuffles. Under draw-3, 81.0% of deals beat both rule-based bots. At least one of our three bots won 59.9% of draw-1 deals and 20.3% of draw-3 deals. That's a floor for these bots, not the game's true winnability.
Now the limits. These are bots, not people. The planner's rules are ours, not a published strategy, and the random bot is a straw man. The rules are Windows-style, not any app's, and nothing here measures Solitaire Real Cash or any other game. What the test does show cleanly: hold the cards fixed, change the decisions or the draw rule, and results move a lot.
Atay Games bot test, September 2026. Three bots played the same 20,000 hidden-card Klondike deals. Under draw-1 rules, a random-move bot won 26.7%, a published rule-follower 42.5% and our strategy-rule planner 51.2%. Under draw-3, the same planner won only 16.3%. Same cards, different decisions, different results; different rules, different luck.
Why is FreeCell so much more skill-based than Klondike?
Because nothing is hidden. FreeCell deals all 52 cards face up, and in 2026 Blake and Gent calculated that 99.998881% ± 0.000207% of random deals can be solved (Blake and Gent, arXiv v6). Call it 99.999%. When a FreeCell game goes wrong, the cause is almost always a decision, not a hidden card.
The famous exceptions are rare. Michael Keller's long-running FreeCell FAQ notes that "All of the 32,000 Microsoft deals except for number 11982 are solvable." Among the first million deals in the FreeCell Pro program, it adds, "Eight of those one million are impossible."
Yet people still lose plenty of FreeCell games. On Solitaired, one large solitaire site, players won only 40.9% of 296,733 FreeCell games (Solitaired, 2024 data). The site itself cautions: "Keep in mind these are human win rates." When almost every deal is solvable and every card is visible, most of that gap is decisions. Some of it may be games people abandon, since the site doesn't say how unfinished games count.
Winnable vs won: the gap is decisions plus hidden information
Winnable = every card known (Blake and Gent, Journal of Artificial Intelligence Research, 2026). Won = games played on Solitaired (Solitaired, "The Odds of Winning Solitaire," dated February 24, 2024); the site doesn't say how abandoned games count, and its rules may differ from the paper's.
Put the three games side by side, and you get a luck dial:
- FreeCell: every card visible. Close to all skill, though the deal still sets how hard a game is.
- Klondike draw-1: many cards hidden, but the whole stock is within reach on every pass. Decisions dominate over many games.
- Klondike draw-3: many cards hidden, and only every third stock card within reach on a pass. Luck-heavy in any single game.
FreeCell sits at the skill end. All 52 cards are dealt face up, and 99.999% of random FreeCell deals can be solved (Blake and Gent, 2026). Of Microsoft's original 32,000 deals, only number 11982 is known to be impossible. Yet players on one large solitaire site won just 40.9% of FreeCell games, a gap that mostly comes down to decisions.
Does a same-deal cash match remove the luck?
No. It equalizes the deal between players, which is a different thing. Skillz, operated by Firy Inc (NYSE: FIRY, formerly SKLZ), runs cash tournaments for many mobile games, ours included. Its help center says that in asynchronous games, "both players play the same match with the same conditions, but they do not have to play at the same time" (Skillz help center).
The same article adds: "Any variables from Player 2's match are recreated for Player 1." That's Skillz's description of its platform as a whole.
Its developer documentation is more careful. The shared random generator gives game developers "the means to ensure that each competitor in a match receives the exact sequence of random values." The same page warns that this "does not guarantee that all players will have an identical game experience," because "Over time, the player's decisions may diverge" (Skillz Random and Fairness).
In Klondike, your choices decide which face-down cards you turn, and when. Two players on one deal can make different guesses and uncover different cards at different moments. Casual play asks, "Can I beat this deal?" A same-deal match asks, "Can I outscore someone on this deal?" The second question strips out the deal's luck between players. It can't strip out the guesses.
What did our same-deal score race show?
The better bot won more often when both bots played the same deal. We ran a simplified score race: more cards on the foundations wins, and if level, fewer moves wins. That's our rule, not any app's formula. Our planner beat the random bot in 77.6% of same-deal draw-1 races, against 71.1% when each played its own shuffle.
| Matchup | Draw-1: same deal | Draw-1: different deals | Draw-3: same deal | Draw-3: different deals |
|---|---|---|---|---|
| Planner beats Random | 77.6% | 71.1% | 72.0% | 67.3% |
| Rule-follower beats Random | 71.2% | 65.1% | 70.2% | 65.7% |
| Planner beats Rule-follower | 57.3% | 56.1% | 54.1% | 52.0% |
| Rule-follower vs its own copy | 49.8% | 50.0% | 49.2% | 49.9% |
| Equal copies finish level on cards (Rule-follower / Planner) | 64.8% / 81.7% | 19.5% / 27.3% | 66.8% / 79.6% | 8.0% / 8.3% |
The mirror rows tell the rest. Two copies of the same bot, differing only in how they broke ties, finished level on cards in 64.8% to 81.7% of same-deal races, against 8.0% to 27.3% on different deals. So the deal's luck was shared. Yet on the rest of those identical deals, roughly one in five to one in three, equal players ended on different card counts, because their choices uncovered different cards.
For the format itself, here's how cash solitaire tournaments work. For tactics, see how to score higher in a solitaire tournament. And because skill ratings pair you with similar players, the gaps you face in real matches tend to be small.
Same deal, not the same game. A same-deal match equalizes the deal; it doesn't remove luck. Skillz says players in an asynchronous match "play the same match with the same conditions." Its developer docs add that shared random values do "not guarantee that all players will have an identical game experience." In Klondike, choices change which cards each player uncovers.
Is solitaire legally a game of skill?
It depends on where you are, and we found no court decision on solitaire itself. US states use different tests. The best known comes from California's Supreme Court in 1962. The court wrote: "The test is not whether the game contains an element of chance or an element of skill but which of them is the dominating factor in determining the result of the game" (In re Allen, 59 Cal.2d 5).
This section is general information, not legal advice.
The tests fall roughly into three groups:
- Dominant factor. Does skill or chance dominate the result? That's the In re Allen test above.
- Material element. Skillz's legal overview describes this test as asking "whether chance plays a material role in determining a game's outcome."
- Any chance. The strictest test. As the law firm Klein Moynihan Turco summarized it in 2018, "If chance influences the outcome of a game in any way, it will be deemed illegal gambling" (Klein Moynihan Turco).
Where would solitaire land under these tests?
Our evidence says the answer shifts with the rules. Draw-1 and FreeCell look decision-heavy over many games, and draw-3 looks luck-heavy per game. But we couldn't find a court that has weighed that evidence for solitaire, so read any page that says "solitaire is legal because it's skill" with care.
Operators know all this. Skillz says in its 2024 annual report that it uses "proprietary algorithms and data science tools designed to ensure that the degree of skill involved in affecting the outcome of a contest is sufficient to comply with applicable state laws." It also says: "We have not received any licenses, authorizations or approvals confirming that the paid entry-fee contests hosted on our platform comply with applicable laws" (Skillz Form 10-K for 2024).
Prize play is 18+ and depends on where you live, so check whether prize play is available in your state. For the bigger picture, here's how the predominance test separates skill games from gambling.
No solitaire ruling, three tests. No court ruling on solitaire turned up in our search. Many US states ask whether skill or chance is the "dominating factor," the test California's Supreme Court set out in 1962 (In re Allen). Others ask whether chance plays a material role, and a few treat any chance as enough. Where you live decides it.
What's the strongest case that solitaire is mostly luck?
For draw-3 Klondike, it's a strong case. Even perfect play with every card known loses about 18% of deals (Blake and Gent, 2026). The best published hidden-card AI loses about 63% (Bjarnason et al., 2009). And on one large site, players won just 11.1% of draw-3 games (Solitaired, 2024 data).
Our own test adds to it. Under draw-3, 81.0% of our 20,000 deals beat both rule-based bots. For a single game, the deal often matters more than the player.
Small skill gaps also look like coin flips. In our same-deal score race, the planner beat the rule-follower only 57.3% of the time under draw-1 and 54.1% under draw-3, even though it won clearly more games outright. Two decent players on the same deal are close to even.
Rated matchmaking pairs similar players, so real single matches feel random, too. We covered why luck wins single matches but not seasons in our gin post, and the same arithmetic applies here.
What cuts against the luck case?
Three things. FreeCell shows that when nothing is hidden, 99.999% of deals can be solved. Decisions multiply win rates on identical deals, in the 2009 study and in ours. And same-deal formats remove the deal gap between players, even though they can't remove the guesses.
So what's our verdict?
The evidence supports this much: decisions predominate over many games in draw-1 Klondike and FreeCell, while luck weighs heavily on any single game of draw-3. Nobody has measured a validated skill percentage for solitaire, so any "70/30" split you read is a guess. For contrast, here's a game the law treats as chance: bingo.
The honest counter-case. The luck case is strong for draw-3 Klondike. Even perfect play with every card known loses about 18% of deals, and the best published hidden-card AI still loses about 63% of games. On one large site, players won just 11.1% of draw-3 games. For a single game, the deal often matters more than the player.
Frequently Asked Questions
What percentage of solitaire games can be won?
It depends on the rules and on what you can see. If you knew every card, 81.9% of draw-3 Klondike deals and 90.5% of draw-1 deals could be won (Blake and Gent, 2026). With cards hidden, the best published AI won about 37% of draw-3 games (Bjarnason and colleagues, 2009).
Is draw-3 solitaire harder than draw-1?
Yes, much harder. With every card known, winnable deals drop from 90.5% under draw-1 to 81.9% under draw-3 (Blake and Gent, 2026). In our 2026 bot test on identical deals, the same strategy bot won 51.2% under draw-1 but only 16.3% under draw-3, about two-thirds fewer wins.
Is FreeCell all skill?
Close to it, but not entirely. All 52 cards are dealt face up, and about 99.999% of random deals can be solved (Blake and Gent, 2026). Of Microsoft's original 32,000 deals, only number 11982 is known to be impossible. The deal still sets how hard a game is, so it isn't luck-free.
Does everyone get the same deal in cash solitaire?
On Skillz, the platform says players in an asynchronous match "play the same match with the same conditions." That equalizes the deal, not all the luck: once your choices differ, you uncover different cards at different moments. Skillz's own developer docs say shared random values don't guarantee an identical game.
Is playing solitaire for money gambling?
It depends on where you live. Many US states ask whether skill or chance is the dominating factor; others ask whether chance plays a material role, and a few count any chance. We found no court ruling on solitaire itself. Real-money play is 18+, so check your state first. This isn't legal advice.
Why do I lose so many solitaire games if skill matters?
Because many deals can't be won by anyone who can't see the hidden cards, and draw-3 rules make it worse. On one large solitaire site, players won 33.0% of draw-1 Klondike games and 11.1% of draw-3 games (Solitaired). Judge your play over dozens of games, not one.
The bottom line
Solitaire is a skill game with a luck dial, and the rules and the hidden cards set it. FreeCell sits near the skill end. Draw-1 Klondike rewards good decisions over many games. Draw-3 Klondike is luck-heavy in any single game. And the 82% stat you'll see quoted describes a game you can't play: draw-3 with every card known.
Since 2018, May 22 has been National Solitaire Day, the date Microsoft chose to mark Solitaire's arrival in Windows (Xbox Wire). The game went mainstream after it shipped with Windows 3.0 in 1990. You don't need to wait for May to feel the dial for yourself, though: play a few draw-1 games, then a few draw-3 games, and compare.
One honest note on money. A skill edge against matched opponents is worth a few dollars, not an income, and entry fees can be lost. Treat paid play as a hobby with a budget.
Want to test your own decisions? Practice free first, then decide whether a cash match in Solitaire Real Cash is for you. It's on iPhone and iPad through the App Store, and on Android as an APK from the official Atay site. Here's why real-money games aren't on Google Play. Real-money play is 18+, and prize availability depends on your location.
Sources
- Atay Games, Klondike bot test: 20,000 deals, seed 20260923, three bots plus two equal-skill copies, draw-1 and draw-3, 200,000 games in total, standard-library Python 3.14.4. Run 2026-09-23. First-party data from a model, not from players; method described in this post.
- Blake, C. and Gent, I., "The Winnability of Klondike Solitaire and Many Other Patience Games," Journal of Artificial Intelligence Research 85 (published 2026-02-27); arXiv 1906.12314 v6 (2026-03-03), abstract, rules text and Tables 1 and 3. Retrieved 2026-09-23. doi.org; arxiv.org. Earlier versions of the research reported slightly different intervals; this post uses the JAIR figures.
- Yan, X., Diaconis, P., Rusmevichientong, P. and Van Roy, B., "Solitaire: Man Versus Machine," Advances in Neural Information Processing Systems 17 (NIPS 2004). Retrieved 2026-09-23. web.stanford.edu
- Bjarnason, R., Fern, A. and Tadepalli, P., "Lower Bounding Klondike Solitaire with Monte-Carlo Planning," Proceedings of the 19th International Conference on Automated Planning and Scheduling (ICAPS 2009), pp. 26 to 33, published 2009-10-16. Retrieved 2026-09-23. doi.org; PDF at ojs.aaai.org
- Solitaired (Neal Taparia), "The Odds of Winning Solitaire, Backed with Data," page dated 2024-02-24: Klondike turn 1, 955,805 wins in 2,898,974 games; turn 3, 158,382 in 1,429,916; FreeCell, 121,239 in 296,733. Retrieved 2026-09-23. solitaired.com
- Michael Keller, FreeCell FAQ, Solitaire Laboratory (undated page). Retrieved 2026-09-23. solitairelaboratory.com
- Xbox Wire (Microsoft), Paul Jensen, "Celebrating National Solitaire Day on May 22," 2018-05-15. Retrieved 2026-09-23. news.xbox.com
- Skillz help center, "How does Skillz player matching work?", article 211525983 (edited 2025-12-02, page updated 2026-07-02). Retrieved 2026-09-23. support.skillz.com
- Skillz developer documentation, "Skillz Random and Fairness" (docs version 2025.0.50). Retrieved 2026-09-24. docs.skillz.com
- Skillz developer documentation, "The Legality of Skill Gaming" (docs version 2025.0.50). Retrieved 2026-09-23. docs.skillz.com
- David Klein, "Games of Skill v. Games of Chance: The Legal Analysis," Klein Moynihan Turco, 2018-08-16. Retrieved 2026-09-23. kleinmoynihan.com
- Supreme Court of California, In re Allen, 59 Cal.2d 5 (1962), via Stanford SCOCAL. Retrieved 2026-09-24. scocal.stanford.edu
- Skillz Inc., Annual Report on Form 10-K for fiscal year 2024 (accession 0001801661-25-000050), SEC EDGAR. Retrieved 2026-09-22. sec.gov
- Apple App Store, Solitaire Real Cash listing (id6473332617, listed as "Solitaire Win Real Cash Skillz"). Retrieved 2026-09-23. apps.apple.com
Legal disclaimer. This article is general information, not legal advice. It summarizes research, platform documentation and one court decision as of September 2026, and the law on paid play differs by state and changes over time. Nothing here is a statement about whether any game, platform or account is lawful where you live. Our bot test is a model, not a measurement of any app or of real players. In cash play, better decisions raise your odds over many games but never guarantee a single result, and you can lose entry fees. Real-money play is 18+ and isn't available everywhere.
Practice Free, Then Decide
Try Solitaire Real Cash in free practice first. If you enter a cash match later, pick an entry fee you're comfortable losing. Cash play is optional, skill-based and 18+.
Play Solitaire Real Cash
