GTO poker: what game theory optimal strategy is and how to use it

GTO poker, short for game theory optimal, means playing a strategy that sits in a Nash equilibrium: a strategy your opponent cannot beat by changing theirs. Heads-up, a player using it cannot lose in the long run against any opponent, however that opponent plays. It is not the strategy that wins the most. Against a weak player who calls too much or folds too much, an adjusted, exploitative strategy earns more.

In practice “GTO” is shorthand for the answers poker solvers produce: how often to bet, which hands to bluff with, how often to call. This guide explains what equilibrium means, shows it on a three-card toy game you can check by hand, gives the numbers behind balanced betting, and covers where GTO stops working and what sites allow during play.

What GTO actually means

Game theory studies decisions where your best move depends on what someone else does. Poker fits perfectly: whether a bluff is good depends on how often your opponent calls, and whether a call is good depends on how often you bluff.

Nash equilibrium in plain words

A Nash equilibrium is a pair of strategies where neither player can gain by changing theirs alone. Rock-paper-scissors is the classic example. If you play each option one-third of the time at random, no opponent strategy can beat you on average. If you lean toward rock, a thinking opponent plays more paper and wins. The equilibrium is the mix that leaves nothing to exploit.

Poker works the same way, only with far more options. A GTO strategy for a spot says how often to bet, check, call, raise or fold with each hand, so that the opponent has no profitable change left.

Unexploitable, but only heads-up

The guarantee that an equilibrium strategy cannot lose holds in two-player zero-sum games, which is what a heads-up pot is. With three or more players it disappears. An equilibrium still exists, but playing it no longer guarantees you break even, because two opponents’ mistakes can help one another at your expense. The researchers behind Pluribus, the program that beat professionals at six-max No-Limit Hold’em in 2019, said so directly: their bot did not play an equilibrium and could not have been sure of winning by playing one.

Why GTO is not the most profitable strategy

Equilibrium is a defence. It makes sure you cannot be exploited, but it also does not go out of its way to exploit anyone. If an opponent calls far too often, the most profitable answer is to stop bluffing them and bet your value hands bigger. A GTO strategy keeps bluffing at its balanced rate and loses those extra chips. Against weak players, GTO wins, but less than a well-aimed adjustment.

The AKQ game: GTO in a single hand

Full poker is too big to solve by hand, but a toy game shows every idea in a few lines. This one, the AKQ game, is a classic of poker theory. Work through it once and the logic of bluffing frequencies and bluff-catching stops being mysterious.

The rules: the deck has three cards, an ace, a king and a queen. Each player antes 1, so the pot is 2, and each gets one card. The first player, the bettor, may check or bet 1. If they bet, the second player, the caller, can call or fold. If the bettor checks, or the bet is called, the higher card wins the pot.

Some decisions are easy. The bettor with the ace always bets, because it always wins. The bettor with the king gains nothing from betting: an ace calls and wins, a queen folds and would have lost anyway, so the king checks. The caller with the ace always calls, and with the queen always folds, since a queen can never beat a bet. The whole game comes down to two questions: how often does the bettor bluff with the queen, and how often does the caller call with the king?

The answer is one in three for both. Here is why each number holds the other in place.

• The bettor’s bluffing rate. When the bettor bets, it holds an ace in 3 cases out of 4 and a queen in 1 out of 4. A king that calls risks 1 to win 3 (the pot of 2 plus the bet), so it needs to win 1 time in 4 to break even. With bluffs at exactly a quarter of bets, calling and folding are worth the same to the king.
• The caller’s calling rate. A queen’s bluff risks 1 to win 2, so it breaks even if it is called 2 times in 3. Against a queen, the caller holds an ace half the time (always calls) and a king half the time. If the king calls 1 time in 3, the bluff is called ½ + ½ × ⅓ = ⅔ of the time. Bluffing and checking are now worth the same to the queen.

Neither player can do better by changing their number. If the queen bluffed more, the king should call every time; if it bluffed less, the king should never call. If the king called more, the queen should never bluff; if it called less, the queen should always bluff. Only the one-in-three pair leaves nothing to exploit. Played this way, the bettor wins on average about 3% of the pot more than an even share, simply because it is the one allowed to bet.

At equilibrium, every hand that sometimes takes two different actions must be indifferent between them. That is why the king calls only as often as it takes to make the queen’s bluff break even.

The numbers behind balanced betting

The AKQ logic works for any bet size, and it gives two numbers that matter in real hands. The first is how much of your betting range can be bluffs, so that a bluff-catcher is indifferent to calling. The second is minimum defence frequency (MDF): how much of their range the opponent must continue with, so that your pure bluffs with any two cards do not show an automatic profit.

The pattern is clear on a chart: bigger bets let you bluff more, and they force the opponent to fold more of their range.

Two numbers in the table look alike but are different, and it helps to see them in one hand. Say you are on the button with Q♥ J♥ and miss everything on a K♥ 9♣ 4♥ 2♠ 3♦ board. The big blind checks the river into a $20 pot, and you bet $15.

Your opponent would call $15 to win $35, so they need 30% equity. That means a balanced betting range here is 30% bluffs, 3 bluffs for every 7 value bets. Your bluff on its own is a different question: it risks $15 to win $20, so it needs the opponent to fold 15 ÷ 35 = 42.9% of the time to break even. If you have no idea how your opponent plays, the first number tells you how often to bluff; if you know they fold too much, the second tells you every bluff wins. Defending against this kind of bet is the subject of our guide to minimum defence frequency.

Mixed strategies and indifference

Solver output often shows the same hand doing two things: bet 40%, check 60%. That is not indecision. As in the AKQ game, a mixed hand is one that earns the same with either action at equilibrium, and the exact mix is what keeps the opponent from exploiting the rest of your range.

Which hands end up mixed depends on the spot. Usually they are hands on the border between two groups: the weakest value bets, the best of the bluffs, or bluff-catchers that are just good enough to call. Hands that block your opponent’s strongest holdings are often chosen as bluffs, and hands that block their bluffs as folds; our guide to blockers covers how that works. For a human, the practical lesson is simple: when two actions are close, it matters far less which one you choose than that your overall range is not lopsided.

From Nash to solvers: a short history

GTO is often described as a recent invention. The mathematics is almost a century old; only the computing power to apply it to real poker is new.

John von Neumann proved the minimax theorem in 1928, the basis of equilibrium in two-player games, and his 1944 book with Oskar Morgenstern already analysed a simplified form of poker. John Nash showed in 1950 that every finite game has an equilibrium. In 2006 Bill Chen and Jerrod Ankenman’s The Mathematics of Poker solved toy games such as AKQ and brought the language to players.

Then computers took over. In 2015 the University of Alberta’s Cepheus essentially solved heads-up Limit Hold’em: its strategy is so close to equilibrium that a human could not beat it over a lifetime of play. In January 2017 Carnegie Mellon’s Libratus beat four top professionals at heads-up No-Limit Hold’em over 120,000 hands. In 2019 Pluribus won at six-max No-Limit against elite players, without an equilibrium guarantee, as noted above.

How solvers find GTO

A solver does not know poker strategy. It is given a simplified version of one spot and plays it against itself millions of times, adjusting after each pass. The inputs define the problem:

• both players’ ranges at the start of the spot;
• stack and pot sizes;
• a bet tree: which sizes each player may use on each street;
• rake, and in tournaments the payout structure for ICM.

Most solvers use counterfactual regret minimisation (CFR). After each iteration, every decision is nudged toward the actions that would have done better, and the average of all those strategies converges on equilibrium. How close it gets is measured as exploitability, or “Nash distance”: how much a perfect opponent could win against the result, usually as a percentage of the pot. Solutions within about half a percent of the pot are accurate enough for study.

The answer is only as good as the inputs. Give the solver the wrong ranges or a bet tree with sizes nobody uses, and it returns a perfect solution to the wrong question. Newer tools can handle some multiway spots, but at a much higher computing cost and with simpler trees. How to use them well is covered in our guide to studying with poker solvers.

Where GTO breaks down

Solver outputs are exact for the model and approximate for real games. The gaps matter more at the table than in theory.

• Multiway pots. The no-loss guarantee holds only heads-up. With three players, an “equilibrium” can lose money, and most solver work is still done heads-up.
• Rake. A solution built without rake overvalues small, thin edges. Real games take 5% of the pot, and more live.
• ICM in tournaments. Chips are not money near the money bubble, so a chip-EV solution can be wrong; see our guide to the bubble in poker.
• Simplified bet trees. A solver with three bet sizes does not know how to respond to a fourth.
• Real opponents. Almost no one plays close to equilibrium, so the most profitable play is usually a deviation.

Copying solver frequencies without the reasons is a common way to lose money. A 20% bluff frequency taken from one spot is wrong on a different board, at a different stack depth or against a player who never folds. Learn why a solver bets, not only how often.

GTO vs exploitative play

An exploitative strategy is built around a specific opponent’s mistakes: bluff the player who folds too much, value-bet thinly against the one who calls too much. It earns more than GTO against that opponent, but it opens you to counter-exploitation if your read is wrong or the opponent adjusts.

Good players use both. GTO is the baseline, what to do against an unknown or strong opponent, and the reference that tells you what “folding too much” means. Exploitation is the deviation once you have evidence. The trade-offs are compared in detail in exploitative play vs GTO, and the expected-value maths behind both is in our guide to EV in poker.

Do you need GTO at low stakes?

You do not need to memorise solver charts to beat micro stakes, where most opponents make large, obvious mistakes. But a few GTO ideas pay off at every level:

• Bet-size logic. Knowing that a pot-sized bet needs to work 50% of the time as a bluff and allows one bluff per two value bets keeps your betting ranges sensible.
• Defending against bluffs. The MDF table tells you when you are folding too often to big bets, a common leak at every stake.
• Hand selection for bluffs. Using hands with no showdown value and good blockers, rather than random ones.

What you can safely skip: exact mixed frequencies, tiny bet sizes used to balance against strong players, and close preflop spots. Against loose opponents who call too much, the exploit is to bluff less and value-bet more; our guide to playing against calling stations shows how. Against an opponent who folds too much, the opposite, as covered in our guides to bluffing and fold equity; and for spots where you hold a bluff-catcher, see advanced bluff-catching.

Studying GTO vs real-time assistance

Studying with a solver between sessions is allowed everywhere. Using one, or anything that gives strategic advice, while you play is not. The big sites have tightened their rules in 2026:

The details are in our PokerCraft guide for GGPoker’s built-in tools, and in the CoinPoker and PokerStars reviews.

Real-time assistance (RTA), software that tells you the GTO play during a hand, is cheating on every major site. Sites detect it by comparing your decisions with solver output over thousands of hands, and the penalty is usually a closed account and confiscated winnings, sometimes paid out to the players you beat.

GTO poker FAQ