Poker solver AI: what GTO software solves

What is a poker solver?

A poker solver is software that works out a game-theory optimal, or GTO, strategy for a specific spot. Feed it the stack sizes, the pot, the bet-sizing options and the range of hands each side could realistically hold, and it calculates exactly how often to bet, raise, call or fold with every hand in that range, so that no single deviation from the plan can be exploited. That's the whole idea in one sentence: a balanced strategy nobody can profitably attack.

Just as important is what it isn't. A solver doesn't know your opponent's actual cards, it doesn't watch the table, and it isn't running live during a hand at a real table, that's a different, banned category of software covered further down. What it produces is a reference strategy, computed in advance, for a defined situation.

How does a poker solver work?

Most modern solvers lean on an algorithm called counterfactual regret minimisation, CFR for short. Practically, that means the software plays a huge number of hands against a copy of itself inside the defined game tree, and after every hand it nudges each decision slightly toward whichever action it "regretted" not taking more. Do that millions of times and the small nudges converge on a stable equilibrium, a strategy where, if either side deviates on its own, it only makes its own result worse.

That process is heavy computation: real solvers can take hours to run a single deep, multi-street spot with wide ranges on both sides, which is exactly why commercial tools like GTO Wizard sell pre-solved solution databases rather than asking you to run everything from scratch. PokerSnowie takes a slightly different approach, acting more like a virtual coach: it reviews hands you've played and scores each decision against its own solved strategy, rather than just handing you a chart to look up. Both are serious engineering, closer to the game-theory research world than to a slot predictor app.

About this chart

A page explaining solvers shouldn't oversell its own chart, so to be clear about it: this shows a simplified, single-scenario opening range for a generic 100 big blind, six-max cash game, folded to you, at each position. It's built to be directionally right and useful for learning the shape of a range, tight up front, wide on the button, but it isn't a live solver output. A real solve would flex with the exact rake structure, effective stack depths, table tendencies and whatever sizing options were fed into it, and postflop play branches out into thousands of further decisions this page doesn't attempt to cover.

Toggle between seats and the button range comes out at 39.4% of hands, far wider than under the gun. The reason is position itself. Acting last after the flop is a real, measurable advantage, so a hand that's a clear fold under the gun becomes a perfectly good raise on the button, where fewer players are left to wake up with something strong behind you.

Solver, equity calculator or Chen formula?

All three tools live under Poker AI on this site, and they answer different questions. The Chen formula rates one hand in isolation in a fraction of a second, a fast first pass with no opponent involved. The equity calculator runs one specific hand against one specific opponent hand over thousands of simulated boards, answering "who wins this matchup, and how often." A solver sits above both: it reasons about entire ranges on both sides across every remaining betting round, not just one hand versus one hand, which is why it needs serious computing power rather than a quick simulation. For most study sessions the Chen formula and equity tools are plenty; a full solver licence is a different level of commitment.

Is using a poker solver cheating?

Studying with a solver between sessions, the way this page uses one, is completely standard practice among winning players and isn't cheating in any sense, it's homework. The category that gets a player banned is real-time assistance, sometimes called RTA: software running live during an actual hand, telling you what to do based on the cards on screen. Virtually every online poker room explicitly prohibits this, win or lose it's treated as cheating, and getting caught typically means a closed account and forfeited funds, not just a warning. If a page or forum post is selling a "real time poker solver" you run mid-hand, that's the product category to stay well away from, regardless of how it's marketed.

Poker bots are a related but separate problem: fully automated players that make every decision without a human at all. Detection here has moved well past simple timing checks. Vendors like GeoComply build device intelligence into their poker-integrity products specifically to catch a running solver, an RTA overlay or a remote-access tool in the background, not just to watch how a hand plays out, and rooms such as WPT Global have described staffing dedicated integrity teams that judge success by whether a suspected cheat can sustain a winning rate long-term, not by ban counts alone, since a legitimate player who studies solvers heavily will already look statistically unusual to a simple filter.

Solvers and language models in 2026

The research has moved on since solvers first appeared. Academic work on a system called ToolPoker found that general-purpose language models can describe correct, solver-quality poker reasoning in plain English, then take the opposite action anyway, a gap researchers term "knowing-doing." Giving the model live access to a real solver mid-hand closed most of that gap, but that exact setup is real-time assistance by definition, which is precisely why the paper's authors argued against ever commercialising it for actual play. Separately, a week-long event run by PokerBattle.ai in late 2025 had several language models play each other directly with no tool access at all; OpenAI's o3 finished on top, ahead of Claude Sonnet 4.5, with a wide enough gap to the rest of the field to show these models still play with noticeably different risk profiles rather than one obvious "best" approach. Neither result changes the basic rule on this page: a solver you consult between hands is study, a model or solver feeding you decisions live is not.

Does GTO guarantee winning poker?

GTO is an unexploitable baseline, not a maximum-profit strategy against a specific, weaker opponent sitting across from you right now. Most winning players study solvers precisely so they know what balanced looks like, then deliberately deviate from it to punish real mistakes they see at the table, folding tighter against a player who never bluffs, bluffing more against one who folds too much, and drifting back toward balance against tougher opposition. Memorising a chart also doesn't fix execution: sizing tells, timing patterns and weak postflop decisions cost far more in practice than a slightly imperfect preflop range ever will.

How to study with one

Solvers are study tools, and they work best when you use them that way.

• Review hands after a session, never during one. Real-time assistance is banned on almost every real-money site.
• Build the basics first with the Chen formula and equity calculator.
• For tournaments, add ICM, since a chip-EV solution can be wrong near the money.

Sources and further reading

• Wikipedia: Nash equilibrium
• Wikipedia: computer poker players
• GTO Wizard (solver software)