Rules, captures, and the threats that win five in a row with a twist
Pente is a five-in-a-row game with a rule borrowed from the Go family: you can capture. It was devised by Gary Gabrel in the 1970s as an adaptation of the Japanese game Ninuki-renju, and its name is the Greek word for five. Two players take turns placing stones on the intersections of a 19x19 grid, and there are two ways to win — line up five stones, or capture five pairs of your opponent's stones.
That second route changes the character of the game. In Gomoku a stone, once played, is permanent. In Pente a carelessly placed pair can vanish, taking a half-built line with it, and a stone used to block a threat can be lifted off to reopen it. You have to read two kinds of threat at once, and many games are decided by a capture that breaks a line rather than by the line itself.
One difference from tournament play: competitive Pente makes the first player start on the centre point and place their second stone at least three intersections from it. This version has no opening restriction. If the board fills with no winner, the game is a draw. Picking a difficulty starts a fresh game, and there is no undo.
Whenever two of your stones sit side by side in a line, check both ends. If one end holds an enemy stone and the other is empty (O X X _), your opponent captures by playing the empty end. If both ends are empty (_ X X _), the pair is safe for now, but one enemy stone at either end turns it into a capture threat that can be cashed next turn. Most lost captures come from the same move: extending a single stone into a pair directly beside an enemy stone, which hands over a threat for free. A stone can be part of a safe three horizontally and a vulnerable pair diagonally at the same time.
Because only pairs can be captured, the gap shape X _ X is sturdier than X X. Neither stone can be taken as part of that line, and filling the gap later gives a solid three. The split three (X X _ X or X _ X X) is just as useful: it threatens to become four, it is harder to spot than a solid three, and the gap gives your opponent no safe bracketing point. Likewise, turning a threatened pair into a three removes the threat along that line, because a sandwich of three is not a capture.
Pente players call an open three a tria and four in a row a tessera. The logic mirrors Gomoku: an open tria (_XXX_) must be answered, or it becomes an open tessera that threatens five at both ends and cannot be stopped with one block. Pente adds a second way to defend. A capture that removes even one stone from a line breaks it, and the two stones of a capture always lie along the capture direction — so to break a horizontal four, look for a vertical or diagonal pair that includes one of its stones. If that pair is flanked on one side by your stone, playing the other side takes it and the four is gone without a block.
Captures count toward victory even when they break nothing, and the race sharpens as the numbers climb. Once a player holds eight stones, every capture threat they make is a threat to win the game and must be treated like a four. That creates a powerful double threat: one stone that both extends a line to an open three and brackets one end of an enemy pair. Your opponent can stop the line or save the pair, not both. When the opponent is the one on eight, every pair you own is a liability, so play shapes with gaps.
A block is only as permanent as the stone that makes it. Suppose your opponent blocks one end of your three. If that blocking stone forms a pair with one of their earlier stones, and you already have a stone beyond the pair, you can capture both — removing the block, scoring two stones and reopening your line in one move.
The PlayVersusAI Pente engine follows a fixed order on every level. If it can win at once, by making five or by taking the pair that brings it to ten, it does, even if you also have a four. If you threaten to win next move, it picks the reply that leaves you the fewest winning points, which may be a block or a capture that breaks your line. Only then does it judge the position, telling open twos, threes and fours from blocked ones, spotting split shapes and double threats, and weighing captures, capture threats and the risk of forming a pair that can be bracketed.
Its blind spot is length: long plans stay invisible until they are close.
| Difficulty | What to expect |
|---|---|
| Easy | Takes wins and stops immediate losses, but weighs your other threats at half strength and adds heavy randomness, so open threes and loose pairs often slip past it. |
| Medium | The full one-move evaluation with almost no randomness. It blocks open threes, protects threatened pairs and punishes pairs left beside its stones. |
| Hard | Adds an alpha-beta search, usually three to six moves deep over its dozen best candidates, in about a second at most. It finds double threats and capture combinations, and sees yours coming. |
| AI Training | The same search with no randomness, but it keeps deepening for up to about 1.2 seconds a move and uses a partly finished deeper search when its first choice has been fully checked. In 72 test games against Hard it won 50. |
Free to play - no account needed