When someone asks "is poker solved," the reaction from a long-time player like me is mixed: part excited by the progress of artificial intelligence, part skeptical about what "solved" really means outside of academic definitions. I remember a late-night online cash session years ago where an opponent's line felt disturbingly robotic — tight preflop, oddly fearless on river bluffs. That moment captures the heart of the question: computers have made enormous strides, but has the game we play and love been fully decoded?
What does “solved” mean in poker?
In game theory, a game is “solved” if an algorithm can produce a strategy that is unexploitable: no opponent can consistently earn a profit against it. For perfect-information games like chess or checkers, that definition is clean. Poker, however, is an imperfect-information, stochastic, multi-agent game with hidden cards, betting structure, and real human psychology. So we must separate three layers:
- Weakly solved: For some restricted variants, an algorithm finds a strategy that plays optimally from the starting position.
- Approximately solved: Algorithms compute near-equilibrium strategies with low exploitability but not perfect optimality.
- Practically solved: The strategy is so robust that human players cannot exploit it in meaningful ways in real play.
Thus, when we ask "is poker solved," we need to ask: which variant? Heads-up limit? No-limit? Multiway games? The short answer: some variants are essentially solved; the full, unrestricted game humans play online and in casinos is not.
Milestones in poker solving
Over the past decade a sequence of breakthroughs reshaped our understanding:
- Cepheus (heads-up limit Texas Hold’em) — Researchers produced an essentially unexploitable strategy for heads-up limit hold’em. In practical terms, Cepheus plays so close to game-theoretic optimum that an exploitative edge against it is statistically negligible.
- DeepStack and Libratus (heads-up no-limit) — These systems used advances in counterfactual regret minimization (CFR) and real-time abstraction/refinement to beat professional players in heads-up no-limit matches. They didn't "solve" the entire game, but they demonstrated how approximate equilibrium strategies can dominate top humans.
- Pluribus (multi-player no-limit) — In 2019, Pluribus showed that AI could beat pros in six-player no-limit Texas Hold’em, a massive leap because multiway games are far more complex than heads-up play.
Those are impressive, but notice the pattern: researchers targeted specific formats (heads-up, short-handed, limit/no-limit) and used computational tricks to approximate equilibria. The general, fully open-ended game remains computationally intractable to “solve” in the strictest sense.
Why some variants became solvable
Two factors allowed progress:
- Reduced decision space: Heads-up limit has a small set of betting sizes and only two players, shrinking the state space dramatically compared to multiway no-limit games.
- Algorithmic advances: Techniques like CFR, neural network-guided value estimation, real-time subgame solving, and abstraction/refinement made it feasible to compute near-equilibrium strategies within limited resource budgets.
In plain English: change the rules to remove complexity, throw in smarter search and approximation, and you can reach strategies that feel “solved.” But real-world poker, with many players, diverse stack sizes, and a buffet of bet sizes, balloons in complexity.
Technical snapshot: how modern solvers work
At their core, modern poker AIs combine two broad ideas. First, they reduce the vast game into a manageable model by grouping similar situations (abstraction) or by deciding on a small set of possible bet sizes. Second, they use iterative self-play methods — variants of regret minimization — to converge toward a strategy that minimizes exploitability.
When the AI reaches the real table, it often performs “re-solving” in real time: it models the specific subgame that unfolds and recomputes strategies for that context using a depth-limited lookahead and neural evaluations. This hybrid of offline strategy and online re-solving is what allowed systems such as DeepStack and Libratus to outplay professionals despite not having a perfect game solution.
So, is poker solved for everyday players?
From a practical standpoint, no. Here’s why that distinction matters:
- Most solvers target very specific formats. Online cash games and tournaments span many formats, table sizes, and stack depths; a solver tuned for heads-up play won't automatically translate its dominance to six-max or full-ring games.
- Exploitability metrics are technical — a solver might have minuscule theoretical exploitability but still make moves that are confusing and exploitable by humans at different time scales and styles.
- Human factors matter. Reads, timing tells, betting patterns, and psychological pressure alter the game in ways that pure game-theoretic agents don’t experience. Real-world poker is not just about playing correct GTO (game-theoretic optimal) lines; it’s about adjusting to opponents and extracting value.
Put simply: while parts of poker are "solved" in academic terms, the full messy game — especially multi-player no-limit formats — is not solved in a way that ends human competitive play.
What these advances mean for strategy and learning
If you want to improve, understanding what solvers teach us is practical and actionable:
- GTO is a baseline, not a blueprint. Solvers show balanced frequencies and why certain lines protect you from being exploited. Learning these ranges improves your defense and makes you harder to target.
- Exploitative adjustments still win money. A purely GTO opponent will be unexploitable, but most humans deviate. Good players combine solver-informed baselines with exploitative deviations when they detect tendencies.
- Bet sizing shapes ranges. One of the technical lessons of solver play is that bet size is a strategic tool for range construction, not just a lever for pot control.
When I began using solver-based concepts years ago, I stopped treating certain board textures and bet sizes as “one size fits all.” Small adjustments — folding in marginal spots, increasing aggression on particular turn cards — made a measurable difference in my win rate. Those are the real-world benefits of AI progress.
Ethics, fairness, and online play
The existence of superhuman solvers raises concerns about fairness in online poker. There's a difference between players using study tools to improve and real-time assistance that suggests plays during hands. Many sites ban real-time engine assistance for ethical reasons and to protect the integrity of play.
If you're studying with solvers, use them as a coach, not a cheat. Over-reliance on automated outputs without understanding why a line works can atrophy your decision-making — and at regulated sites, using live assistance can result in bans or legal consequences.
A short list of important papers and systems (for further reading)
- Cepheus — essential results on heads-up limit hold’em.
- DeepStack — a breakthrough in approximate equilibrium for heads-up no-limit.
- Libratus — prominent system that defeated top pros in heads-up no-limit play.
- Pluribus — first system to reliably beat professionals in six-player no-limit hold’em.
Reading these works gives insight into how boundaries of solvability shift with algorithmic and hardware improvements.
Future outlook: will poker ever be completely solved?
Complete, exact solutions for full-scale poker variants are unlikely in the near term because of computational limits and the combinatorial explosion of possibilities. That said, progress will continue along these axes:
- Better abstraction techniques that preserve strategic nuance while reducing computation.
- Hybrid models combining deep learning, search, and domain knowledge to handle larger, more realistic game models.
- Practical tools for players that translate solver insights into accessible training aids rather than raw strategy dumps.
So while "is poker solved" remains a provocative question, the more meaningful one for players is: what can I learn from these advances to play better, ethically and sustainably?
Practical takeaways for players
- Use solver concepts to build a sound baseline strategy, but remain flexible and observant at the table.
- Practice exploitative adjustments responsibly — identify consistent tendencies and adapt your ranges.
- Focus on bet sizing discipline and range construction rather than rote memorization of lines.
- Avoid live assistance during play. Study with tools off-table to develop intuition and judgment.
If you want to explore perspective and tools, check out keywords for casual play and community discussions — they offer accessible ways to try variants and understand how different formats change strategy dynamics.
Conclusion
The honest answer to "is poker solved" is nuanced: several important variants are effectively solved or can be approximated to a high degree, and AI has demonstrated superhuman play in targeted contexts. But the full spectrum of poker — especially multi-player, variable-stakes, human-driven games — remains an open field where skill, psychology, and creativity matter. As a player, the most productive stance is curiosity: learn from solvers, practice disciplined exploitative play, and keep developing the human skills that machines still can't fully replace.
For players who want to experience different formats, test strategies, or join communities, resources like keywords can provide a hands-on environment to apply what you've learned. Poker isn't finished; it's evolving — and that evolution is part of the game's enduring appeal.
