Rummy — the basic game, played properly

This is basic rummy — not 500 rummy, and not gin

The first thing almost every rummy page online gets wrong is which game it is. Pagat says it out loud on the very page these rules come from: "many people use the name Rummy to refer to the game called 500 Rummy on this site, in which more than one card can be taken from the discard pile, and points are scored for cards melded." Those are two different games with two opposite strategies, and calling both of them Rummy is how you end up learning the wrong one.

So, plainly: this table plays the basic game. You take exactly one card from the pile, never a run of them. You score nothing at all for the cards you melded — the winner scores the cards still stuck in everybody else's hands and nothing else. And melds go face up on the table as you make them, where anybody may add to them, which is the part that makes rummy a different game from gin rummy rather than a simpler one.

If you want the game where nothing is melded until somebody knocks and the score is a difference of deadwoods, that is gin rummy, and we have built it too. The two engines share nothing but the way a card is written down.

How to play rummy

The deal

Rummy is played with an ordinary 52-card pack. Two players get ten cards each; three or four get seven each — those are Pagat's sizes and they are what the Players picker on the start screen switches between. The dealer deals one at a time starting with the player to their left, then turns the next card face up to start the discard pile; the rest of the pack is the stock, face down. The player to the dealer's left goes first, so with two players the non-dealer leads.

The draw

Your turn begins with exactly one card: the top of the stock, or the top of the discard pile. One card, not several — taking several is 500 rummy. Taking the face-up card tells every other player something about your hand, and that is the entire information channel in this game, so it is not a free choice.

One rule that catches everybody: you may not throw away the card you just picked up off the pile on the same turn. You may take it now and throw it two turns later, and a card drawn from the stock can be thrown immediately — but the pile card is frozen for that turn. This table simply will not let you do it.

The meld

A set is three or four cards of the same rank — 7♠ 7♥ 7♦. A run is three or more consecutive cards in one suit — 4♣ 5♣ 6♣ 7♣. In the standard game the ace is low: A-2-3 is a run and Q-K-A is not.

Once you have drawn, you may put one meld face up on the table. One combination per turn is the standard rule, and it is a real constraint — a hand holding two melds takes two turns to get them both down, and the second turn may never come. (Some people play that you may lay down any number; that is a variant and it is in the engine, though not wired to the picker.)

Get them down. This is the strategic heart of the game and it is the opposite of gin rummy. A meld on the table is worth zero to you — but it is also worth zero against you, permanently, from the moment it leaves your hand. Sit on a made run because you do not want to give the others somewhere to lay off, and the first time somebody else goes out you get caught holding your whole hand instead of three loose cards. We measured exactly that; the numbers are further down and they are brutal.

Laying off

After melding you may add cards to any meld already on the table — yours or anybody else's — and there is no limit to how many you may lay off in one turn. A fourth card of the rank joins a set; a card at either end joins a run, and it chains, so laying a 9♥ onto 10♥-J♥-Q♥ opens the slot your 8♥ needs.

What you may not do is rearrange an existing meld. Pagat's own example: if 2♦ 2♣ 2♥ 2♠ and 3♦ 4♦ 5♦ are both down, you cannot move the 2♦ out of the group into the sequence so as to lay off the A♦. This engine enforces that structurally — there is no operation anywhere in it that moves a card already on the table, so the rule cannot be broken by a bug.

Going out

You go out by getting rid of every card in your hand, by melding, by laying off, or by discarding — a final discard is not required in the standard game. (Some people require it; that variant is implemented too.)

And here is the half that is always missed: the moment somebody goes out, play stops. "There can be no further melding or laying off, even if the other players have valid combinations in their hands." A loser holding a completed run counts it in full, at face value, like any other card. Clones let the losers dump their melds first, which quietly makes the game much gentler than it is.

Going rummy is emptying your whole hand in a single turn having never melded or laid off earlier in the deal, and it doubles the score for the hand. Under the one-meld-per-turn rule you can still do it — but only if everything you are not melding lays off onto melds that are already down. That interaction is not obvious and it is one of the things this engine's test suite pins by name.

The count

The player who goes out scores the total value of the cards in everybody else's hands: face cards 10, aces 1, everything else its face value. Nothing for their own melds. Play continues, hand after hand, until somebody reaches the agreed target.

How long is a game? We measured ours rather than guessing, and it changed the default. Pagat says the target is agreed before play, and this is exactly why it matters. Against Easy, which throws good cards away, the hauls are big and a race to 100 finished in seven or eight hands. Against Medium, where both sides meld on sight and lay off everything they can, the loser is caught with almost nothing — hands came out at three to eight points, and even a race to 50 ran 13 to 19 hands. A four-handed game reaches 100 in about eight, because three losers pay at once.

So the target is a picker on the start screen and the default here is 50, not 100. That is a measurement rather than a rule: 100 is the figure most rule books suggest and it is one click away, but a race to 100 against a bot that plays properly is a very long evening, and we would rather show you the numbers than ship a game nobody finishes.

If nobody goes out — because the stock ran out for the last time, or because you are playing Block Rummy — then everybody counts their hand, the lowest hand wins, and it scores the sum of the differences between its count and each other player's. Pagat gives the worked example and our suite asserts it verbatim: A has 6, B has 15, C has 7, D has 21; A wins and scores 25 = 9 + 1 + 15. A tie for lowest shares the winnings.

When the stock runs out

The discard pile is turned over without shuffling to form a new stock — the whole pile, which is what Pagat's text says; keeping the top card back as a new pile is a different game by one card. Pagat suggests play ends when the stock is exhausted for the third time, and that is the default here. Block Rummy, on the Rules picker, never reuses the pile at all: the first time the stock runs out, the hand is counted where it stands.

Exactly which rules this table plays, and where they fork

Every rule that the published sources genuinely disagree about is a switch in the engine rather than a decision baked in and called "the rules". Here is the whole list, with the default this page uses, so nothing has to be guessed at.

One shortfall, named. When several arrangements of a hand tie for the minimum leftover value, the engine returns the canonical one. A human might prefer a tied arrangement that gives the opponents fewer places to lay off; this engine does not, and does not pretend to. It can only matter when two arrangements score identically.

Play a friend, or a stranger, in a four-character room

Press Private room and you get a four-character code and a link. Send either one to whoever you want to play and they are sitting opposite you — no account, no app, no email address on either side, nothing to install. Find an opponent drops you into the oldest rummy room that is waiting; if nobody is waiting it opens one for you and seats the next person who comes looking. Live rummy is two players, ten cards each, basic rules, and a race to 50 — the same measured target as the bot games, because there is no negotiate-the-target handshake in a room and a live match to 100 ran eighteen hands and was still going when we tested it. Live games count toward a rating; bot games do not, because a browser can lie about its own score and we would rather label that than launder it.

Characters, not letters, and that is deliberate. A code looks like JUX5 or PQ7M — four picks from ABCDEFGHJKMNPQRSTUVWXYZ23456789, which is the alphabet with every lookalike left out: no I, no L, no O, no 0, no 1. That is 923,521 codes, and none of them can be mistyped by somebody reading one out down a phone. Every other game on this site used to call it a four-letter code, which would have told you a perfectly good code like JUX5 was invalid; all thirty-three of those sentences were corrected the day this page was written.

A dropped connection does not end the game. Refresh, close the laptop, lose signal in a tunnel — the same link puts you back in your own seat with the table, the score and the history exactly where you left them. While you are away the house bot covers your turns so your opponent is not staring at a dead table, and it hands the seat straight back when you return.

No engine help against a person. The coach is refused outright in a live game — in the game logic, not merely hidden in the interface. Showing you the best arrangement of your own cards is not engine help; it is arithmetic you would do by hand at a kitchen table, and it is on screen in both modes.

The deal, and why you do not have to trust us about it

Every card game online gets the same accusation and gives the same non-answer: players say the deal turns against them, operators reply that their random number generator is secure. Both can be true at once, which is exactly the problem — a secure generator says nothing about the code that decides when to call it.

So before a card is dealt this page publishes SHA-256 of a secret seed in the panel beside the table, and you supply a seed of your own. Every shuffle is a pure function of those two values and the hand number and nothing else — not the score, not who is winning. Each hand's seed is released the moment that hand is scored, the master seed when the game ends, and Check the deals re-derives every deal in front of you from the published numbers alone. In a live game both seats commit before either one opens, so neither side can go fishing for a pair of seeds it likes.

And what it does not prove, because a hash that reassures without proving anything is worse than no hash at all. Against the bots the whole thing runs in your own tab — there is no server, so this proves the deal was fixed before the first card and matches the seed, and it proves nothing about a machine somewhere else. In a live game one seat derives the deck, which means that browser holds the seed while the hand is played: the protocol proves nobody chose your cards, not that a modified opponent client could not look at them. Hiding a hand from the computer that dealt it is mental poker and we have not built it. The same protocol on our dice carries a SHA-256 self-test against the published FIPS 180-4 vectors, and it is the same hash function this page uses.

The arithmetic underneath, and the published numbers that check it

Rummy's hard problem is not the rules. It is one optimisation, asked on every card you touch: what is the least this hand can be worth if I arrange it as well as possible? A card can serve a set or a run and never both, which is what makes the obvious answers wrong. So the engine ships two independently written solvers — one a bitmask interval dynamic program over four 13-bit rank masks, the other a brute enumeration of every legal meld and every disjoint subset of them — and the suite makes them agree.

⭐ But the anchor here is external and it is reached exhaustively. Three or four players are dealt seven cards each, and the seven-card hand is one of the most heavily published objects in combinatorics — the seven-card poker frequency table. A poker "four of a kind" is exactly a rummy set of four, and a poker "straight flush" is exactly a rummy run of five or more in one suit. So the published integers are a direct test of this engine's geometry, and the suite walks all 133,784,560 seven-card hands — no sample, no reduced deck — and reproduces every one:

🔑 And the anchor has teeth on the one rule this game argues about. Poker's straight flush counts the ace both ways; basic rummy's ace is low only. So 41,584 is reachable only with the ace-high-and-low setting, and the default must give a different number — 37,444. A run detector that quietly allowed Q-K-A would sail through a "does A-2-3 meld" unit test and fail the published integer. The suite asserts both directions.

What is not published, said plainly. There is no published table of minimum-unmelded-value over rummy hands, with or without jokers. We looked: OEIS has no rummy sequence, Pagat, Bicycle and Hoyle are rule books, and the only academic treatment of rummy proper — Christopher Finkle's 2017 Stetson University paper on expected waiting times — studies invented three- and four-card variants and publishes waiting times, not a value distribution. That is why the optimiser is anchored by two independent solvers agreeing rather than by a citation, and why the ace geometry both of them depend on is pinned from outside by the poker figures.

The two solvers also found a genuine fork in Pagat's wording and this engine resolves it in public rather than silently. Pagat says a joker "can be used as a substitute for any card needed to meld a set or run", and "any card" turns out to be ambiguous: may a joker impersonate the 7♦ while you are holding the real 7♦ in another meld? This engine says no — use the real card if you have it — for two reasons: nobody at a table lets you lay a joker as a card that is face up in front of you, and allowing it breaks the joker-swap rule, because two melds could then each claim the same card and "the card that a melded joker represents" stops being a single card. The permissive reading is a genuinely harder optimisation problem and it is not implemented.

What the bots do, and the idea that lost every single game

None of them search ahead, none of them count cards, and none of them model you. One-ply evaluation with one real idea each. Every win rate below was measured over hundreds of games with the seats swapped, so a seat advantage cannot be read as strength.

🔴 The first Hard bot written for this game lost 100% of its games — 0 wins of 120 — and the reason is the most important strategic fact about rummy, so it is recorded rather than quietly deleted. The idea was imported from gin rummy, where it is correct: hold your melds back, because every meld on the table is a place the opponent can lay off. It still went out 316 times in 840 hands, so it was not stuck. It went out plenty and lost every game anyway.

Basic rummy counts the cards in your hand. A meld on the table is worth zero to you and zero against you, permanently, the moment it is down. Gin rummy counts unmelded cards whether or not they are on a table, so holding them costs nothing there. Here, withholding means that every time the opponent goes out first you are caught with your entire hand instead of three loose cards, and the occasional cheap win does not come close to paying for it. A bot ported from a gin engine will play exactly this way and lose. So will a person.

Easy melds the moment it can and throws a loose card at random. Medium melds at once and throws whichever card leaves the least value behind. Hard is Medium plus the one classic rummy skill: keeping what is nearly a meld. Medium will happily break up a pair or a two-card run to save a single point; Hard scores each discard as the value left behind minus a weight times the live partnerships still standing in the rest of the hand.

⭐ That weight was measured, not chosen. Hard versus Medium, 600 games per cell, both seats, same seeds:

🔑 The top of that curve is flat and the error bars overlap. 1, 2 and 4 are not distinguishable at 600 games, so 2 is "somewhere in the middle of the good region", not a proven optimum, and it is not claimed as one. What is outside the noise is that 0 loses and 8 falls back: letting the potential term overrule the count entirely is worse than ignoring it.

And two ideas that were tried and are worth nothing, recorded because a bot page that lists only the wins is advertising. Shed the biggest meld first (under the one-meld-per-turn rule, put down the four rather than the three): 48.8% ± 4.0 on its own — a tie, and if anything on the wrong side of it. It is still in the code because it costs nothing and reads sensibly, but it is not why Hard wins. Do not feed the table — penalise throwing a card that lays off onto a meld already down — moved the win rate by not a single game at any weight from 0 to 8, because the policy lays off everything it can before discarding, so by the time a card is a discard candidate it is already known not to fit any meld. The term was always zero. A tuning sweep that returns identical numbers at every weight is measuring a dead input, not a flat optimum, and that is the tell.

Nothing here is sold

The rummy apps that dominate the mobile stores are coin economies: you get chips, you lose chips, and then the tables are shut until you buy more or watch a video. So, plainly: there are no coins here, no chips, no buy-in, no energy, no daily limit and no advert between you and the table. All three bot levels, the Learn coach, live play against people and unlimited hands, with no account on either side. There is nothing to buy on this page.

To be fair to the free web rummy games we are actually judged against — cardgames.io and a handful of others are genuinely free, need no account and have no coins. They are not the ones the coin complaints are about, and we would rather say so than let a complaint about a coin economy do work against a site that has none. What they do not do, and what this page is built around, is tell you what your hand is currently costing you, sort it into its exact best arrangement on every card you draw, explain which of the many published rummy variants it is playing, or give you any way to check that the deal was not chosen.

Questions people ask

How do you play rummy?

Ten cards each with two players, seven with three or four. Draw one card — the top of the stock or the top of the discard pile — then you may lay one meld on the table and add cards to melds already down, and then you throw one card away. A meld is three or four of a rank, or three or more in sequence in one suit. Empty your hand and you score everybody else's leftovers. First to 100 wins.

What is the difference between rummy and gin rummy?

In rummy the melds go on the table as you make them, anyone may lay off onto them, and the winner scores the losers' remaining cards. In gin rummy nothing is melded until somebody knocks and the score is the difference of the two deadwoods. The right strategy is opposite: in rummy, get melds down immediately.

Is rummy the same as 500 rummy?

No. In 500 rummy you may take several cards from the discard pile and you score for the cards you melded. Here you take one, and you score nothing for your own melds.

Can you lay off onto other players' melds?

Yes, any number of cards in one turn, onto anybody's melds. You may not rearrange an existing meld to make room.

Is the ace high or low?

Low in the standard game — A-2-3 is a run, Q-K-A is not, and the ace is worth 1. The ace-high variant makes Q-K-A legal and the ace worth 15, because those two halves travel together.

How is rummy scored?

The player who goes out scores the value of every card left in the other hands: face cards 10, aces 1, the rest face value. Nothing for their own melds. Going rummy — clearing your whole hand in one turn having never melded before — doubles it.

How long does a game take?

Longer than you would think, and we measured it. Heads-up hands are worth only three to eight points when both sides play well, so even a race to 50 ran 13 to 19 hands against our Medium bot; a four-handed game reaches 100 in about eight. That is why the default target on this page is 50 rather than the 100 the rule books suggest — 100 is one click away on the start screen.

What happens when the stock runs out?

The whole discard pile is turned over without shuffling and becomes the new stock. Play ends on the third exhaustion, and then the lowest hand wins the sum of the differences. Block Rummy never recycles at all.

Can I play rummy online with a friend?

Yes — Private room gives you a four-character code and a link, and whoever you send it to is at the table. No account, no app, no email on either side. Find an opponent pairs you with a stranger instead.

Is this rummy free?

Yes. Three bot levels, a Learn coach, live play, every hand, no account, no coins, no chips, no buy-in, no daily limit, no adverts between you and the table.