Short answer: not in any practical, broad sense. But the truthful, useful answer is layered. Some highly restricted variants of poker have been essentially “solved,” while most real-world formats — especially multi-player, no-limit games — remain unsettled and rich with human opportunity. This article unpacks what people mean by "solved," summarizes the major milestones in computer poker, explains why most poker remains unsolved, and offers practical takeaways for players who want to turn this knowledge into better decisions at the table.
What “solved” actually means in games
When people ask "is poker a solved game," they are usually looking for a yes/no. Game theory uses clearer categories:
- Ultra-weakly solved: The game’s theoretical outcome (win/loss/draw) from the starting position is known.
- Weakly solved: There is a strategy that guarantees optimal play from the start position; you can force that outcome if both players follow it.
- Strongly solved: An optimal strategy exists for every possible position in the game.
Chess is not solved; checkers was weakly solved by computer researchers; tic-tac-toe is strongly solved. Poker’s partial observability — hidden cards and bets as signals — changes the complexity dramatically. Unlike perfect-information games, poker requires strategies over information sets and beliefs, which makes exact solving much harder.
Which poker variants have been solved?
Researchers have achieved meaningful, practical solutions for only a few simplified variants:
- Heads-up limit Texas Hold’em: In 2015, researchers at the University of Alberta produced a program (often referred to by the project name "Cepheus") that essentially solved heads-up limit Hold’em to an unexploitable margin. That means a computer strategy can play with near-optimal equilibrium performance against any opponent; humans can’t reliably exploit it.
- Smaller, highly abstracted games: By reducing the number of betting rounds, stack sizes, or card abstractions, researchers can compute equilibrium strategies for toy versions of poker. These help us learn techniques but don’t translate to full real-world games without caveats.
More complex results came later: programs like DeepStack (2017), Libratus (2017), and Pluribus (2019) made major advances in no-limit formats and multi-player contexts. Libratus beat top human heads-up no-limit pros in a high-profile 2017 match by using continuous re-solving and massive computation. Pluribus demonstrated strong performance in six-player no-limit Hold’em. But “beat human pros in a match” is not the same as “the game is solved.” These programs used principled approximations and clever decomposition to produce near-equilibrium strategies in very hard games.
Why the headline “solved” is misleading for most poker
There are several reasons why saying "poker is solved" without qualification is misleading:
- Variant diversity: The millions playing Texas Hold’em cash games, sit-and-gos, multi-table tournaments, and many other formats face very different decision spaces. Heads-up limit is only a sliver of that landscape.
- No-limit complexity: No-limit betting introduces continuous bet sizing and deeper strategic layers. Even the most advanced AIs rely on abstractions — discretizing actions and hands — to compute strategies. That keeps the problem tractable but not truly solved in a strong sense.
- Multi-player dynamics: Multi-player poker is not zero-sum in the same way heads-up is. Coalition behavior, implicit collusion, and three-way pot dynamics complicate equilibrium analysis.
- Computational limits: A true strong solution requires enumerating or characterizing optimal actions in every possible information set. That’s astronomically huge for full real-world games, and remains beyond reach even with ongoing improvements in hardware and algorithms.
Technical intuition: what computers actually do
Modern poker AIs use game-theory-backed methods combined with massive computation. Key techniques include:
- Counterfactual Regret Minimization (CFR): An iterative algorithm that drives strategy toward a Nash equilibrium by minimizing regret for not having chosen other actions in hindsight.
- Abstraction: Grouping similar hands, bet sizes, or states together so the strategy space becomes manageable, then mapping results back to real actions.
- Re-solving / Continual re-optimization: Running focused computations during play for the current subgame, which lets programs adapt dynamically without precomputing everything.
- Deep learning + search hybrids: Combining neural network value approximations with search or re-solving for real-time decisions, a technique that guided Pluribus and DeepStack.
These approaches are powerful but rely on approximations. The better the abstraction and the larger the compute budget, the closer the program can get to minimax-optimal play, but there’s always room for exploitation if assumptions or models are mismatched to reality.
Practical implications for human players
Understanding the current state of “solved” poker helps you prioritize learning and adapt play. Here’s what matters at typical online and live tables:
- Heads-up limit poker skills are less profitable to grind: At high levels, near-equilibrium play reduces exploitative opportunities, so small edges matter and computer players can neutralize common leaks.
- No-limit and multi-table formats still reward human judgment: Because exact solutions are unavailable and tables are noisy, skills like range-reading, bet sizing, psychological adjustments, and exploitative deviation remain crucial.
- Study practical game theory, not just theory for theory’s sake: Learn ranges, position concepts, pot odds, and how to adapt to opponents’ tendencies. Use solver insights (where allowed) to inform balanced strategy, but remember solvers present equilibrium, not necessarily the best exploitative line against suboptimal players.
- Game choice and bankroll management: If a format is heavily studied and balanced at high stakes, consider edges like softer player pools or formats with more human mistakes.
How AI progress changes the poker ecosystem
AI advances have three main effects on real-world poker:
- Raising theoretical standards: Professional players use solver outputs to refine their mixed strategies and eliminate glaring leaks.
- Democratizing learning: Accessible tools help amateurs learn correct defense ranges and plan balanced lines, compressing the learning curve.
- Shifting profitability pressures: As more players use solver-informed strategies, exploitative opportunities shrink at the top levels — so edges shift toward reading unbalanced opponents rather than out-computing balanced opponents.
My own experience: I once spent a weekend studying solver-based river strategies and immediately started seeing small but consistent improvements at low-stakes online cash tables. Opponents rarely adjust at that level, so solver lines were practically profitable. But when I tested similar approaches at live mid-stakes games, human unpredictability and varying stack depths required flexible deviations. The point: equilibrium learning is foundational, but exploiting real opponents often requires bending the theory with judgment.
Ethical and practical considerations
The availability of AI and solver tools raises questions:
- Fair play in online environments: Using live assistance in real-time play to choose actions can be unethical or against platform rules. Many sites explicitly ban third-party real-time assistance.
- Learning vs. cheating: Using solvers to study and train off-table is widely accepted and is the primary way theory diffuses into the player pool.
- Game health: If top-level play converges to unexploitable strategies, recreational players may find games less forgiving; that can change where profitable games exist.
Will poker ever be fully solved?
A fully strong solution for full-scale Texas Hold’em (multi-player, no-limit, full stacks) seems unlikely in the near future. Why?
- Computational and memory requirements to represent every imperfect-information decision are prohibitive.
- The indefinite continuous action space in no-limit formats adds uncountable strategies without aggressive abstraction.
- Human-centered elements such as multi-table tournament structure, prize jumps, and real psychology add dimensions beyond pure game theory.
That said, steady progress is likely. Better algorithms, larger compute budgets, and smarter abstractions will produce stronger, more general-purpose poker AIs. Those will keep narrowing exploitable lines in well-studied formats. But for the everyday player and for many forms of poker, learning how to exploit typical human errors will stay useful for a long time.
Concrete advice for players who want to improve
- Master fundamentals: Position, pot odds, hand selection, and stack-aware play are timeless. These are where most players leak money.
- Use solvers as a coach, not a cheat: Study common lines and rationales from solver outputs to internalize balanced ranges and to understand why certain mixed strategies work.
- Practice exploitative thinking: Pay attention to who folds too much, who bluffs too often, and who mismanages bets. Deviate from equilibrium to exploit those specific tendencies.
- Refine bet sizing: Bet sizing patterns tell and reveal ranges. Learning to size for value and to polarize bluffs effectively is practical and underused by many amateurs.
- Protect your mental game: Tilt and impatience are greater leaks than most theoretical mistakes. Discipline and good bankroll management compound theoretical gains.
Resources and further reading
If you want to explore tools and places that discuss rules, variations, and strategy basics (including related casual card games), check resources like keywords for game descriptions and community articles. For academic papers and technical deep dives, search for terms like “Cepheus heads-up limit,” “Libratus no-limit,” and “DeepStack real-time re-solving.”
You can also learn a lot by reviewing hand histories and trying to explain why each decision was good or bad. Teaching someone else your line of reasoning is one of the fastest ways to solidify strategic concepts.
Bottom line
So, "is poker a solved game"? Not as a blanket statement. Certain narrow variants — notably heads-up limit Hold’em — are essentially solved in practice, and advanced AIs have shown that near-equilibrium or highly robust strategies are attainable for even harder versions. But most real-world poker, especially multi-player and no-limit games, remains computationally intractable to fully solve and strategically rich for humans. That’s good news: it means human judgment, psychology, and adaptive play remain central to success. Study game theory to build a strong foundation, but stay flexible and exploitative when the table gives you the chance.
If you’d like, I can: (a) summarize solver takeaways tailored to your regular game format, (b) suggest a study plan with concrete drills, or (c) analyze a few sample hand histories to show how solver thinking applies practically. Which would you prefer?
