Is Gin Rummy a Game of Skill or Luck? The Evidence

We gave two computer players the exact same shuffled decks, 100,000 games' worth, and changed one rule in how one of them played. The bot allowed to pick up a useful face-up card won 91% of the games. The other one still won about three hands in ten. That's gin rummy in two numbers: luck is real in any single hand, and decisions win over time.

Home players often swear the game is luck. Cash-app players hear that it's all skill. Both are half right, and neither side usually shows its work.

So we've tried to. Below, you'll find where the luck in gin rummy actually lives and what same-deck AI research found. Then come our own experiment, what courts have really said about card games, and the strongest case that it's mostly luck.

So, is gin rummy a game of skill or luck?

Both, and not equally. Luck from the deal and the stock swings single hands, while draw, discard and knock decisions repeat all game and decide results over many games. That's also the question courts ask. In 1962, California's Supreme Court said the test is which factor dominates, not whether chance exists at all (In re Allen, 59 Cal.2d 5).

In the court's words: "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." Every card game has some chance in it. The useful question is how much the choices move the outcome.

Lawyers call this the predominance test. It's one of the tests states use, and it sits beside the three elements of gambling: consideration, chance and a prize. We explain how the predominance test separates skill games from gambling in its own guide, so this post sticks to the evidence.

Not all evidence is equal, so we ranked it. Most pages that answer this question lean on the bottom two rows of this table: opinion and folklore.

The short verdict. Gin rummy is a game of skill with real short-run luck. In 1962, California's Supreme Court framed the legal test as "which of them is the dominating factor in determining the result of the game." Gin's recurring decisions dominate over many games, but no study has put a validated percentage on gin's skill share.

Where does the luck in gin rummy actually come from?

Three places: the ten cards you're dealt, the order of the stock, and what your opponent throws. The deal matters least. In Atay Games' 2026 simulation of 2,000,000 random deals, only 1 hand in 1,600 was ready to knock straight off the deal. And 69.3% started with 50 or more points of deadwood.

Atay Games simulation, 2,000,000 random ten-card deals, September 2026. Only 39.8% of dealt hands hold even one meld.

So almost every hand is built in the draws, and the draws are where the decisions live: which pile to take from, which card to throw, when to knock. The full deadwood distribution from 2 million deals is charted in our rules guide.

The stock order is another story. It's pure luck, and it can hand one player gin by the fourth draw no matter how well the other one plays. Nobody controls it. What a player does control is how many of those cards they can use, and how few useful ones they give away.

Scoring adds noise of its own. In a constructed example in our scoring guide, one player wins six of nine hands and still loses the game, because points decide it, not hand counts. You can see that nine-hand scoresheet here. Counting hands won is a poor way to judge who played better.

Where the luck lives. In a 2026 Atay Games simulation of 2,000,000 random ten-card deals, only 0.0625% of hands (1 in 1,600) were dealt at 10 deadwood or less. Just 0.0013% (1 in 80,000) were dealt gin. The average dealt hand carried 56.5 points of deadwood. Gin rummy hands are decided in the draws, not the deal.

How much do a player's decisions change the result?

A lot, and it's measurable. In 2021, researchers Larkin, Collicott and Hiebel reported at the AAAI conference that their heuristic agent, Tonic, won "against the baseline agent at least 68% of the time" (AAAI-21). In their tests, that was 68% against a baseline that knocks at 4 deadwood or less, and 60% against a baseline given Tonic's own knocking rule. Both sides were dealt at random by the same game engine. Only the decisions differed.

That baseline is weak on purpose. Its author, Todd Neller of Gettysburg College, describes it in the challenge's public code as "a random dummy Gin Rummy player that has the following trivial, poor play policy." It ignores what its opponent does, discards whichever card leaves it the least deadwood and knocks at the first chance (EAAI Gin Rummy code, GitHub).

How often the better agent won, on randomly dealt games

Source: Larkin, Collicott and Hiebel, "Modeling Expert Knowledge in a Heuristic-Based Gin Rummy Agent," AAAI-21, 2021. The paper's abstract summarizes the result as "at least 68%" against the baseline agent.

Two facts belong side by side here. Better decisions beat that baseline about two games in three. Yet the better agent still lost about a third of its games to a version of that poor player set to knock at 4 or less. And against one using Tonic's own knocking rule, it lost 40%. Those losses are a rough gauge of how much luck survives a real skill gap.

Which decisions carry the edge? That's a strategy question, and our advanced gin rummy strategy guide covers knock timing, discard safety and pile counting. This post only needs to show that decisions matter. The next section shows how much.

Same kind of deals, better decisions, more wins. In 2021, Larkin, Collicott and Hiebel reported at AAAI that their heuristic gin rummy agent, Tonic, beat the EAAI challenge's baseline player at least 68% of the time. Against a baseline using Tonic's own knocking rule, it won 60%. Even so, it lost about a third of its games to a knock-at-4 version of a baseline whose author calls its play policy "trivial, poor."

What happened when we gave bots identical decks?

On identical decks, decisions won. In September 2026, Atay Games ran 100,000 games for each of three bot pairings. Our port of the challenge's baseline bot beat a bot that discards at random in all 100,000 games. It also beat a copy of itself missing one rule in 91.4% of games.

What did the results show?

The random bot never won a game. It won 1,059 of 269,070 decided hands, or 0.39%. That's roughly what luck alone did for a player making no real decisions.

The one-rule gap is the more telling result. The weaker copy plays exactly like the baseline, except it never picks up the face-up card, even when that card would finish a meld. Without that one habit, it lost 91.4% of games (95% confidence interval 91.2% to 91.5%), where an identical copy loses about half. Yet it still won 31.5% of decided hands, and 8.6% of games, about 1 in 12.

Games and hands won by the baseline bot, on identical decks

Atay Games bot test, September 22, 2026. First-party data; method above.

Why is the game gap so much wider than the hand gap? Because a game to 100 points adds up many hands, and the stronger bot also won bigger ones. It averaged 109.6 points per game to its weaker copy's 40.3. Small per-hand edges compound, which is the arithmetic of the next section.

The mirror run makes the opposite point, and it matters just as much. When two identical bots swapped the same cards, the same seat won both games in 72.3% of pairs, whichever copy sat in it. If the cards made no difference, that would happen only about half the time. So with skill exactly equal, the cards and the seat did much of the deciding.

Now the limits. These are bots, not people. The baseline is weak, the rules are the research version rather than any app's, and a bot that discards at random is a straw man. What the test does show cleanly is this: hold the cards constant, change the decisions, and the results change a lot.

Atay Games bot test, September 2026. Across 100,000 duplicate-dealt games under EAAI challenge rules, our port of a published baseline gin rummy bot played a copy of itself that never takes the face-up card. It won 91.4% of games (95% CI 91.2% to 91.5%). The weaker copy still won 31.5% of decided hands and 8.6% of games.

Why does luck win single matches but not seasons?

Because a real edge is small in any one match and large over many. By plain binomial arithmetic, a player who truly wins 60% of matches still loses 40% of single matches. After 101 matches, though, they're behind only about 2% of the time. We call that the luck horizon.

The luck horizon: chance a better player is behind after N matches

Binomial arithmetic computed by Atay Games, September 2026. A math illustration for hypothetical players, not observed data. Odd match counts, so there are no ties.

A smaller edge makes the picture harsher. A 55% player is still behind after 101 matches 15.6% of the time. Losing four in a row proves very little either way.

Rated matchmaking makes single results feel even more random. On the standard Elo scale, a 100-point rating gap means about a 64% expected score, and a 200-point gap about 76%. Skill-rated pools keep gaps small, which pulls every edge toward 50%. Skillz, the platform run by Firy Inc (NYSE: FIRY, formerly SKLZ), says in its 2024 annual report that it matches "comparable players, on both skill level and tournament template." Our guide to how skill ratings pair opponents covers the rest.

Luck decides the match; skill decides the ledger. By binomial arithmetic, a player with a true 60% win rate still loses 40% of single matches. That player trails after 11 matches 24.7% of the time, after 51 matches 7.4%, and after 101 matches 2.1%. A 55% player still trails after 101 matches 15.6% of the time.

Does playing the same deck remove the luck?

No. It equalizes the luck of the deal between the two players; it doesn't remove luck itself. Gin Rummy Cash's App Store listing says: "Both players are dealt the same deck and play against the same computer opponent. The player who earns the most points wins!" Your competitor is a real person, and your point totals are compared.

What a same-deck match equalizes

Based on the Gin Rummy Cash App Store listing (id6742690853), checked September 22, 2026. Match length and bonus settings are not shown because we haven't verified them in the app.

What does that equalize? Neither of you can blame the deal. You start from the same cards and face the same computer-controlled opponent inside the hand. What it doesn't equalize is everything after your first different choice. Take a different card than your competitor did, and the cards you each see start to diverge.

Skillz's own documentation says so plainly. Its random generator ensures "that each competitor in a match receives the exact sequence of random values," but that "does not guarantee that all players will have an identical game experience." In its words, "Over time, the player's decisions may diverge and the same random number will not be used in the same place" (Skillz Random and Fairness). That divergence comes from decisions, which is the whole point of the format.

Duplicate bridge exists for the same reason, and bridge is the game California's Supreme Court looked at in 1962. The court accepted "an element of chance resulting from the deal of the cards." It still found bridge "predominantly one of skill," because of "a continually recurring necessity in the bidding and play of the hand to make decisions."

Same-deck play isn't a niche format, either. In 2024 the Skillz platform hosted an average of over 1.1 million daily tournaments, about 405,000 of them paid entry, according to its Form 10-K for 2024.

Same deck, not the same game. Gin Rummy Cash's App Store listing says both players are dealt the same deck and play against the same computer opponent, and the higher score wins. Skillz's fairness documentation adds that identical random sequences do "not guarantee that all players will have an identical game experience," because each player's choices change which cards they see.

What have courts actually said about rummy and card games?

No US appellate decision we could verify rules on gin rummy by name. The closest US authority is In re Allen (California, 1962), which found bridge "predominantly one of skill." The clearest rummy ruling comes from India. In State of Andhra Pradesh v. K. Satyanarayana, India's Supreme Court called rummy "mainly and preponderantly a game of skill."

This section is general information, not legal advice.

The Indian court's reasoning, from its decision of 22 November 1967, is worth reading. Rummy, it said, "requires certain amount of skill because the fall of the cards has to be memorised and the building up of Rummy requires considerable skill in holding and discarding cards." It even tied the point to bridge: "The chance in Rummy is of the same character as the chance in a deal at a game of bridge" (judgment text, CaseMine).

It also added a caveat. If "the owner of the house or the club is making a profit or gain from the game of Rummy or any other game played for stakes, the offence may be brought home." That case concerned rummy as played in an Indian club, not American gin (here's how Indian rummy differs from gin), so treat it as a persuasive analogy rather than precedent.

What about the famous 1965 ruling? According to gin rummy author Pramod Shankar, a federal court in Nevada ruled in 1965 that a gin tournament was a contest of skill. No published case record has surfaced, so treat it as reported history.

Why isn't a skill finding the last word?

Skill findings don't always hold. In United States v. DiCristina, a federal judge in Brooklyn found poker predominantly a game of skill in 2012. The Second Circuit reversed on August 6, 2013. It held that the federal gambling statute covered the poker business either way. In the court's words, "the question of whether skill or chance predominates in poker is inapposite to this appeal" (United States v. DiCristina, 726 F.3d 92).

India just delivered the same lesson. In 2021, the Madras High Court struck down Tamil Nadu's 2021 ban on online games played for stakes, including rummy and poker. On 27 May 2026, India's Supreme Court set that judgment aside in State of Tamil Nadu v. Junglee Games India (2026 INSC 594). The protection given to games of skill, it held, "cannot be construed as protection to betting on games of skill, which essentially becomes a gambling enterprise" (Supreme Court Observer).

India's national law had already gone further. Its Promotion and Regulation of Online Gaming Act, 2025 bans online money games. It defines them as games played for stakes "irrespective of whether such game is based on skill, chance, or both."

The lesson: a skill finding helps, but the statute decides. Lawmakers write the definitions, and they can rewrite them. Pennsylvania added "skill slot machines" to its Gaming Act in 2017, and this year its Supreme Court used that wording to classify skill game machines as slot machines.

What do the operators themselves say?

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 adds: "We have not received any licenses, authorizations or approvals confirming that the paid entry-fee contests hosted on our platform comply with applicable laws."

So a game's skill status rests on the operator's own reading of each state's law, not on a license. Prize play is 18+ and depends on where you are, so check whether prize play is available in your state.

Card-game rulings favor skill; stakes are a separate question. California's Supreme Court found bridge "predominantly one of skill" in 1962, and India's Supreme Court called rummy "mainly and preponderantly a game of skill" in 1967. But the Second Circuit reversed a poker-is-skill ruling in 2013, and in May 2026 India's Supreme Court held that betting on games of skill can still be prohibited.

What's the strongest case that gin rummy is mostly luck?

The best-known way to measure skill points the other way for many games. In 2020, economists Duersch, Lambrecht and Oechssler reported in the European Economic Review that most popular online games, poker included, fall below their 50% skill threshold ("Measuring skill and chance in games"). Nobody has measured gin rummy that way.

Their method ranks players with Elo ratings and asks how spread out the ratings are, compared with chess where half the games are swapped for coin flips. By that yardstick, in the published abstract's words, "poker contains about as much skill as chess when 75% of the chess results are replaced by a coin flip."

Here's the part rummy fans won't like. The paper's 2018 working version also measured an online rummy game from a European gaming site, and it fell below the 50% benchmark too (working paper, Table 2). It wasn't gin, but it's the nearest relative anyone has measured this way.

Even the bridge ruling had a warning in it. Dissenting in In re Allen, Justice McComb wrote that "in other than duplicate bridge the element of chance may well be predominant in many a match, particularly when the opposing players are of somewhat equivalent skills."

Our mirror run puts numbers on that sentence. When two identical bots swapped the same cards, the same seat won both games in 72.3% of pairs, against about 50% if the cards didn't matter. And a clearly weaker bot still won 31.5% of decided hands.

What cuts against the luck case?

Another study leans the other way, with caveats. A 2025 preprint by Banerjee and colleagues, using data on 20,000 players, scored two-player rummy at 0.62 on a skill scale where chess is 1.00 (arXiv:2410.14363). It isn't peer reviewed, the authors call their scores tentative, and the rummy data came from Games24x7, an Indian gaming company.

So here's our verdict. The measured evidence supports "decisions predominate over many games." It doesn't support a precise percentage, and anyone quoting 70/30 or 80/20 for gin is guessing.

The honest counter-case. In 2020, Duersch, Lambrecht and Oechssler found that most popular online games, including poker, fall below a 50% skill threshold on an Elo-based measure (European Economic Review). No one has applied that method to gin rummy, so there is no validated skill percentage for gin, only evidence that decisions shift results over many games.

Frequently Asked Questions

The bottom line

Gin rummy is a game of skill with real short-run luck. The deal almost never settles a hand. Decisions measurably change win rates, even between bots that differ by one habit. Luck still owns single hands and single matches, especially between evenly matched players. And courts have treated decision-heavy card games as skill, but whether you can play for money is a separate legal question. For a game the law defines as chance, see the same question for bingo.

One more honest note. 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 try a $1 match in Gin Rummy Cash. 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. For formats, prizes and cash-outs, see our guide to playing gin rummy for real money.

Legal disclaimer. This article is general information, not legal advice. It summarizes court decisions and research 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. Skill-based cash games involve real money: 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 is not available in every jurisdiction.