GTO Wizard AI Benchmarks

GTO Wizard has combined the power of artificial intelligence with traditional solving methods to bring you high-quality GTO poker solutions at incredible speeds. What used to take hours of computation on high-end hardware can now be done in seconds from your browser!

This fantastic advancement naturally leads us to an exciting inquiry: How precise are these rapid solutions? We benchmarked our solutions against conventional solvers to find out.

How Accurate Are GTO Wizard's Solutions?

Overview

GTO Wizard AI Solutions

Benchmarks against Slumbot
Benchmarks against PioSolver

Speed and accuracy
Exploitability
NEW Engine (November 2024 Onwards)
OLD Engine (Prior to November 2024)
Methodology

Overview

We strive to deliver high-quality GTO strategies. That’s why we’re transparent about the accuracy and solving methods of our solutions. Most GTO Wizard solutions have an accuracy between 0.1% and 0.3% of the pot:

Postflop Accuracy is measured in Nash Distance, which represents the exploitability of the solution, measured as the maximum potential EV loss of the current solution in big blinds divided by the pot. Lower Nash Distance is preferable. To put that into perspective, a 0.3% loss in a 5bb pot would mean that solution could be exploited for, at most, 0.015 bb/hand, or about 1.5 bb/100.

Pre-solved solutions are computed beforehand using standard CFR solvers, such as our General and Simple solutions. Pre-solved solutions are solved to an accuracy of at least 0.3% pot. However, lower SPR spots such as 3BP or 4BP are often solved to much higher accuracy (0.1% pot).
Custom Solutions are solved on the fly according to the parameters you input. These are solved using state-of-the-art technology developed by Ruse AI, which was acquired by GTO Wizard. Nash Distance was measured by nodelocking flop strategies.
All river spots in GTO Wizard are re-solved in real time using a traditional CFR solver to 0.1% Nash Distance.

You can check the accuracy of solutions using the preview icon in the solution selector.

See how ↓

GTO Wizard AI Solutions

GTO Wizard AI Solutions are solved on the fly according to the parameters you input. These are solved using state-of-the-art technology that combines traditional solving methods with predictive neural networks to solve any spot within seconds! But how accurate are these AI-Generated solutions?

We’ve benchmarked these strategies against PioSolver and Slumbot with incredible results!

Benchmarks against Slumbot

Slumbot is an open-source poker bot that plays an approximate Nash EquilibriumNash Equilibrium
A stable set of adversarial strategies where no player can gain by unilaterally changing their own strategy. strategy. When playing, it plays according to a gigantic pre-computed solution and maps its opponent’s observed action and bet size to one or more nearby bet sizes within its abstraction. Slumbot won the most recent Annual Computer Poker Competition (ACPC), making it a powerful nemesis!

GTO Wizard AI beat Slumbot for 19.4 bb/100 in a 150k hand Heads-Up match. In this match, each player was given only 7 seconds to make their move. This technology is way ahead of what can be achieved with any other software!

Benchmarks against PioSolver

The ultimate test of these solutions is to see how well they perform against PioSolver.

Speed and accuracy

Traditional CFR solvers, like PioSolver, cannot solve spots quickly. They require powerful hardware and significant time in order to achieve high accuracy. The chart below compares the speed and accuracy of GTO Wizard and PioSolver:

PioSolver – Required 4862 seconds (1 hour 21 minutes) to reach 0.23% accuracy. Hardware: 16 cores @ 5GHz, 128GB RAM.
GTO Wizard – 6 seconds to reach 0.22% accuracy. Hardware: 2 Cores, 8GB of RAM.

100bb HU SRP – A♠7♥3♦ – 95 GB Tree

Exploitability

We solved 900 flops in order to benchmark our custom solutions against PioSolver!

In order to test how accurate our solutions are, we nodelocked our flop strategies into PioSolver and measured the EV loss.

NEW Engine: November 2024 Onwards

Note – In November 2024 we upgraded our engine. The engine now produces solutions with lower variability, ensuring greater confidence in your strategy adjustments. These improvements mean you can trust that every solution is optimized for maximum accuracy, helping you refine your strategies with greater precision than ever before.

Average Flop Exploitability: 0.165% pot
Most solutions achieved a Nash Distance between 0.09% – 0.24% pot.

The following interactive chart displays how often (vertical ↑) GTO Wizard AI achieves different levels of accuracy (horizontal →).

GTO Analyzer V2: Precision Just Got Better

By The Numbers:

OLD Engine: Prior to November 2024

Show/Hide ↓

Average Flop Exploitability: 0.21% pot
Most solutions achieved a Nash Distance between 0.15% and 0.30% pot.

The following interactive chart displays how often (vertical ↑) GTO Wizard AI achieves different levels of accuracy (horizontal →).

All Spots

By Preflop Action

By Stack Depth

6max vs HU

Methodology

EV losses are shown ± one standard deviation.
Loss vs. Best Counter-strategy – How much EV you’d lose against an Exploitative Strategy, AKA Nash Distance.
Loss vs. Fixed GTO Strategy – How much EV you’d lose against a fixed Nash Equilibrium strategy (GTO).

Here’s how we measured the accuracy of our custom solutions:

We solved 900 postflop spots in PioSolver across various formats and preflop actions, then solved those same spots with GTO Wizard AI for comparison.
We nodelockedNodelocked
The act of fixing one players strategy at some node within a solver. Typically this is done to calculate exploitative adjustments. GTO Wizard’s entire flop strategy into PioSolver for one player and measured the EV loss against the current strategy (Loss vs. Fixed GTO).
Then we re-solved with that nodelock in place, allowing Pio to maximally exploitMaximally Exploitative Strategy (MES)
A strategy that will perfectly maximize value to take advantage of all your opponent’s mistakes. See Nemesis strategy.
GTO Wizard’s AI fixed flop strategy, and measured the EV loss (Loss vs. Best Counter-strategy AKA Nash Distance).

The keen among you may ask: Why did we only nodelock the flop? Why not nodelock turn and river as well? The reason for this is because AI-solvers do not actually calculate the strategy on future streets. Rather, they predict the EV of future streets using neural nets and lookahead functions. After you solve a flop there is no turn strategy to nodelock (that would be a new solve altogether). It should be noted that rivers that are solved with GTO Wizard use traditional CFR (like PioSolver) as these spots are small enough to be solved on the fly.

We are fully committed to delivering our users with unprecedented speed and precision in the market of GTO learning tools. That’s why we are constantly working on the quality of our neural networks and the precision and speed of our results. If you have any questions, feel free to reach out to us on our discord channel!

Check out these articles to learn more about GTO Wizard AI: