Solvers and GTO: A Practical Guide for Modern Poker Players

The emergence of solver-driven game-theory-optimal play over the course of the 2010s represents the most consequential shift in poker strategy since the publication of David Sklansky's Theory of Poker in 1978. It is worth, before addressing the practical question of how a modern player should use solvers in their own study, setting out the history of the underlying mathematics, the tools that have implemented it, and the players who have driven its adoption.

This is the ground I intend to cover in what follows.

The Mathematical Origins

The foundational insight, that a two-player zero-sum game of imperfect information admits a mixed-strategy equilibrium from which neither player can profitably deviate, belongs to John von Neumann, who proved the minimax theorem in 1928 and elaborated it in the Theory of Games and Economic Behavior, co-authored with Oskar Morgenstern and published by Princeton University Press in 1944. Von Neumann's formulation treated gambling, specifically poker, as a motivating example, and the treatment of a simplified bluff-call game in chapter nineteen of that book is the direct ancestor of modern GTO analysis.

The gap between von Neumann's simplified game and the full mathematics of no-limit Texas hold 'em was, however, immense. The 1944 treatment dealt with a game of perhaps a few hundred decision nodes. A single hand of no-limit hold 'em, considered as an extensive-form game, has on the order of 10^160 decision nodes when bet sizes are discretised to a reasonable grid. No direct solution was possible with the computational resources of the mid-twentieth century, nor of the late twentieth.

The breakthrough was algorithmic rather than purely mathematical. In 2007, researchers at the University of Alberta's Computer Poker Research Group, working under the supervision of Michael Bowling and others, developed and published the counterfactual regret minimisation algorithm (CFR). CFR is an iterative self-play algorithm that, over many iterations, converges on an approximate Nash equilibrium of the game. It was capable, with the computational resources of the time, of approximately solving simplified variants of hold 'em (initially heads-up limit hold 'em) to a precision that was, in the language of the field, effectively solved.

In 2015 the Alberta group announced that heads-up limit Texas hold 'em had been solved in the weak sense (a strategy within a statistically detectable distance of the equilibrium had been computed). In 2017 and 2019, the Libratus and Pluribus systems, developed by Tuomas Sandholm and Noam Brown at Carnegie Mellon University, demonstrated superhuman performance against professional heads-up no-limit and six-max no-limit players respectively.

The Emergence of Commercial Solvers

The research tools of the academic groups were not immediately available to players. What changed the landscape for practical players was the development of commercial solvers designed to compute approximate equilibrium strategies for specific situations in no-limit hold 'em.

PioSolver, released by Polish engineer Rafal Kurgel in 2015, was the first such tool to achieve broad adoption. PioSolver computed, given a specified set of preflop ranges, bet sizes, and board textures, an approximate equilibrium strategy for the resulting subgame. The tool was computationally intensive (a single moderately complex spot might take hours to solve on consumer hardware) but was operable by a reasonably technical player without requiring a doctorate in computer science.

Subsequent commercial solvers followed. GTO+ launched in 2017. Simple Postflop, developed by a Russian team, followed at a similar period. MonkerSolver specialised in multi-way pots. By the early 2020s, a solver infrastructure existed through which a committed player could investigate essentially any situation in no-limit hold 'em and determine the equilibrium response.

The Pedagogical Revolution

The availability of solvers did not, by itself, produce a revolution in how poker was taught. That required a second layer: a community of coaches and content producers who could interpret solver outputs and translate them into heuristics a student could carry to the table.

The key figures in this translation effort, across the late 2010s and early 2020s, included Doug Polk (whose Upswing Poker platform became a principal venue), Jonathan Little (whose Pocarr academy likewise), Fedor Holz (through Pokercode), and a cluster of Eastern European theorists including Uri Peleg, Michael Acevedo, and Matthew Janda. Janda's book Applications of No-Limit Hold 'em, published in 2013 and revised subsequently, was an early canonical text. Acevedo's Modern Poker Theory, published by D&B Publishing in 2019, was arguably the first textbook to teach solver-derived strategy in a systematic form.

The pedagogical development followed a recognisable trajectory. Early solver students attempted to memorise specific solver outputs for specific situations. This proved both infeasible (the space of situations is too large) and often counterproductive (the solver's strategy for one board texture does not translate cleanly to a superficially similar one). The field shifted, by the early 2020s, toward extracting principles from solver outputs rather than memorising the outputs themselves. Coaches taught students to notice recurrent patterns in how the solver partitioned its range, sized its bets, and distributed its bluffs. Students learned to anticipate the solver's likely answer in a new situation by applying the principles, then to verify the intuition by running the specific solve.

The Contemporary Practical Picture

For a player in 2026, the practical question is: how should solvers figure in a study routine?

Several considerations bear on the answer.

First, solver study is most valuable against opponents who themselves play close to equilibrium. Against amateur opponents whose strategies deviate materially from equilibrium, the optimal counter-strategy is not the equilibrium itself but a targeted exploit. A player who spends all their study on solvers and none on opponent-specific exploitation is leaving substantial expected value on the table in soft games.

Second, solver outputs are only as reliable as the inputs. A solve that assumes unrealistic preflop ranges, unrealistic bet-size menus, or unrealistic rake structures will produce strategies that are technically optimal for the unrealistic game but that may fail against the actual game. A skilled solver operator is careful about input selection and cross-checks results against multiple specifications.

Third, the memorisation-versus-principles distinction continues to matter. A student who emerges from a solver session with a memorised line for a specific spot is unlikely to apply it well at the table. A student who emerges with a principle (say, that the in-position caller on dry paired boards should develop a substantial check-behind range) has extracted lasting value.

Fourth, the time-allocation question. A serious study routine in 2026 typically allocates perhaps 30 to 50 percent of study hours to solver work, with the remainder divided among hand-history review, opponent-profiling, database analysis (through tools such as PokerTracker and Holdem Manager), and game-selection practice. A pure solver diet is unusual and generally not recommended by experienced coaches.

A Note on the Evolution of the Games

The populations at the online tables have themselves shifted in response to solver availability. A mid-stakes online cash game in 2026 is populated by players who have, as a group, substantial solver exposure. The game is closer to equilibrium than it was in 2015, and considerably closer than it was in 2005.

One consequence, documented in aggregate win-rate data published by PokerTracker and by independent researchers, is that the typical winning player's edge has narrowed. Where a competent mid-stakes regular in 2005 might have expected to win at 5 to 8 big blinds per 100 hands, a competent mid-stakes regular in 2026 is typically looking at edges of 2 to 4 big blinds per 100 hands, with rake and variance consuming a larger fraction of that edge than in the earlier period.

This narrowing is not uniform across games. Soft live environments, niche variants, and the larger multi-table tournament fields where recreational players remain a majority continue to offer larger edges to capable players. The solver revolution has hit hardest in the mid-stakes and high-stakes online cash-game ecosystem that was, in the early 2010s, the principal venue for professional players.

Practical Recommendations for the Modern Student

In closing, and synthesising the above, I offer the following observations to a player beginning a solver-based study routine.

1. Start with principles, not outputs. Read Acevedo's Modern Poker Theory or a comparable text before attempting to run your own solves. The grounding saves enormous time.
2. Run solves in a workflow that includes input cross-checking and principle extraction. A solve session that does not produce a written principle is usually a wasted session.
3. Invest in opponent profiling alongside solver study. The expected value of exploitation against weak opponents typically exceeds the expected value of equilibrium refinement against strong ones.
4. Accept that the technology is not a substitute for hands played. The solver shows you the strategy; you still have to execute it at the table, against a live opponent, with a time limit and the adrenaline of a real pot in front of you.

The availability of solvers has raised the floor of what is possible for a dedicated student of the game. It has not, in my assessment, changed the fundamental character of the game itself. Poker remains a contest of judgment, of psychology, and of temperament, played within the probabilistic structure that von Neumann first described in 1928 and that the Carnegie Mellon and Alberta teams have, over the intervening century, brought steadily into clearer view.

The tools are now here. What remains is the old, human work of learning to use them.