About GTOkiller

What the exploitative solver does, how the engine works, where the population data comes from and who builds it.

GTOkiller is the first exploitative poker solver built on real population data (MDA), not on GTO equilibrium. It models how the player pool of a specific site, stake and player profile actually plays, computes the strategy that maximizes EV against that pool across the full decision tree, and shows exactly which leak each line exploits.

What the tool does

For every spot in a hand, GTOkiller shows three strategies side by side. The exploit is visible, verifiable and auditable: you see what is being attacked, why, and how much EV the leak is worth.

The three-strategy view shown on every decision node
Strategy	What it shows	Source
GTOkiller	The exploitative recommendation that maximizes EV against the pool's real tendencies	GTOkiller exploitative engine
GTO	The theoretical equilibrium strategy a classic solver would give	Classic CFR solver
Population (MDA)	The real frequencies with which the pool plays that spot	Mass Data Analysis of the pool

GTOkiller

What it shows: The exploitative recommendation that maximizes EV against the pool's real tendencies
Source: GTOkiller exploitative engine

GTO

What it shows: The theoretical equilibrium strategy a classic solver would give
Source: Classic CFR solver

Population (MDA)

What it shows: The real frequencies with which the pool plays that spot
Source: Mass Data Analysis of the pool

AI Coach: on demand, in each spot, it explains the recommendation in three steps: the spot and the highest-EV line, the EV breakdown that names the leak with its numbers, and the generalizable lesson. Every statement is anchored to a figure of that node; there is no free-form reasoning
Meta selector: site, stake and opponent profile. Each meta is its own dataset with its own solved tree, because population tendencies are not universal
MDA Explorer: real preflop population ranges by site, stake and position, next to the GTO baseline
REST API: the same engine for bots, trainers and integrations

How the engine works

GTOkiller is a three-layer pipeline: population modelling (MDA), an exploitative solver, and a presentation layer with the AI Coach.

Modelling (MDA): millions of real hands from the target pool are aggregated into a complete decision tree. Each node stores the real frequency of every action, the real range behind each action and the sample size of the node
Exploitative solver: with the opponent fixed to the pool model, the engine optimizes only our side of the tree and maximizes EV while keeping a solid base where the pool's deviation is marginal or counter-adaptation would be easy
Solution and AI Coach: the solved strategy with an EV breakdown per node, presented as the three-strategy view and explained on demand

Key figures of the current pipeline
Parameter	Value
Tree depth	About 500,000 nodes per complete decision tree
Data window	Rolling two years; older hands are discarded
Refresh cadence	Full reprocessing every six months
Sample threshold	Dynamic per spot. Nodes without enough data are discarded, never filled with theory
Profile segmentation	Regulars, discriminated by VPIP and PFR

Tree depth

Value: About 500,000 nodes per complete decision tree

Data window

Value: Rolling two years; older hands are discarded

Refresh cadence

Value: Full reprocessing every six months

Sample threshold

Value: Dynamic per spot. Nodes without enough data are discarded, never filled with theory

Profile segmentation

Value: Regulars, discriminated by VPIP and PFR

If a node does not have enough sample, GTOkiller does not force a solution. Rare spots may show fewer available actions: if the pool only uses one bet size there, only that size is shown, because it is the only one verified.

Why it is not a GTO solver

A GTO solver runs hundreds of iterations of the same hand in which both players improve against each other until they converge to an equilibrium: a strategy that is optimal against itself. It is the perfect answer to the question "how do I play against a clone of me using the same solver". If the opponent plays anything else, which every human does, a superior strategy exists by definition.

GTOkiller solves a different problem: given that the opponent plays as the pool's MDA measures, which strategy maximizes my EV while staying solid enough not to lose money if the opponent adapts reasonably? The opponent is fixed with real data, and only our side is optimized. The output is a solid exploitative strategy: it attacks structural leaks and keeps a balanced base where the exploit would be fragile.

It also solves all streets at once. When a user node-locks one frequency in PioSolver or GTO Wizard, the remaining nodes are still solved as pure GTO, so the result is a partial exploit that loses the synergies between streets. If the pool overfolds on the flop and overbluffs on the turn, the best line may be to check the flop and let them bluff, which a single-node lock cannot see. GTOkiller resolves the whole tree with real frequencies in every node of every street.

GTOkiller compared with GTO solvers and MDA platforms

Solvers compute equilibrium against a perfect opponent. MDA platforms show what the pool does but do not compute the strategy to exploit it. GTOkiller integrates both layers: it reads the pool, computes the line and explains the rationale.

GTOkiller compared with GTO solvers and MDA platforms
	GTO solvers (GTO Wizard, PioSolver, GTO+)	MDA platforms (MDA Demon, Snap MDA, Datadonk)	GTOkiller
Opponent model	Perfect opponent (equilibrium)	Real pool data, no solver	Real pool data as solver input
Output	Balanced strategy	Frequencies and ranges of the pool	EV-maximizing exploitative strategy
Exploiting deviations	Manual node locking, one node at a time	Left to the user	Full tree, all streets simultaneously
Explanation	None or generic	None	AI Coach anchored to the node's numbers
Exploit visibility	Not applicable	Data only	Three-strategy view on every spot

Opponent model

GTO solvers (GTO Wizard, PioSolver, GTO+): Perfect opponent (equilibrium)
MDA platforms (MDA Demon, Snap MDA, Datadonk): Real pool data, no solver
GTOkiller: Real pool data as solver input

Output

GTO solvers (GTO Wizard, PioSolver, GTO+): Balanced strategy
MDA platforms (MDA Demon, Snap MDA, Datadonk): Frequencies and ranges of the pool
GTOkiller: EV-maximizing exploitative strategy

Exploiting deviations

GTO solvers (GTO Wizard, PioSolver, GTO+): Manual node locking, one node at a time
MDA platforms (MDA Demon, Snap MDA, Datadonk): Left to the user
GTOkiller: Full tree, all streets simultaneously

Explanation

GTO solvers (GTO Wizard, PioSolver, GTO+): None or generic
MDA platforms (MDA Demon, Snap MDA, Datadonk): None
GTOkiller: AI Coach anchored to the node's numbers

Exploit visibility

GTO solvers (GTO Wizard, PioSolver, GTO+): Not applicable
MDA platforms (MDA Demon, Snap MDA, Datadonk): Data only
GTOkiller: Three-strategy view on every spot

GTOkiller vs GTO Wizard, in detail · GTO Wizard alternatives in 2026

Where the data comes from

The opponent model is built from aggregated hand histories of the target pool, never from a single player. Each node is estimated from its own sample and only published when the sample is large enough to be trusted. Data is filtered by site, stake and a two-year window, segmented by player profile, and fully reprocessed every six months so the strategies track how the pool evolves.

Plans

The Free plan gives one flop per day with every node, the three-strategy view and the AI Coach. Edge (39 EUR or 49 USD per month) and Killer (79 EUR or 99 USD per month) unlock unlimited study; semiannual plans cost five months.

See all plans and prices

Who is behind it

GTOkiller is built by Advanced Poker Research, S.L., a four-founder company based in Donostia-San Sebastián, Spain, founded in 2025. The team has played online poker since 2011, up to NL1k, and lived the moment when PioSolver and GTO Wizard outputs became the accepted truth of poker study.

From inside the tables the picture was different: no opponent played the way the solver said they should. Entire populations at NL100, NL200 and NL500 deviated in systematic, predictable and measurable ways, while solvers kept computing the optimum against a perfect opponent who never showed up. GTOkiller was born from that gap: if the opponent does not play GTO, a superior strategy exists, and it is the one that exploits their real deviation.

Contact: Questions about the data, the methodology or a partnership: info@gtokiller.com or the Discord community.

Glossary

Exploitative solver: A solver that computes the EV-maximizing strategy against a specific, imperfect opponent model instead of the equilibrium against a perfect one. GTOkiller uses the real pool as that model.
MDA (Mass Data Analysis): The analysis of millions of real hands from a specific population to model its tendencies, frequencies and errors as a statistical model of the pool.
GTO: Game Theory Optimal. The equilibrium strategy computed by algorithms such as CFR: unexploitable, but not EV-maximizing against imperfect opponents.
Leak: A systematic, suboptimal tendency in how a player or a pool plays a spot, such as overfolding to a c-bet or overbluffing the river.
Pool / meta: The set of players at a given site, stake and profile, and the predominant tendencies of that set.
Node locking: Manually fixing a frequency in one node of a GTO solver so the engine solves assuming that behaviour. It exploits one node while the rest of the tree stays GTO.
Three-strategy view: GTOkiller's presentation of the exploitative line, the GTO line and the population's real frequencies side by side on every spot.

Full glossary: 25 terms