Are Nosebleed Players GTO?

It’s no secret to anyone that we live in the GTO era.

The first and most popular GTO solver that appeared in the market was PioSolver, which was launched in the second semester of 2015. Almost 9 years ago!

That’s crazy. It doesn’t seem like it was that long ago.

I caught the early development of this era, as I started taking poker seriously exactly in the end of 2015. I started 2016 playing microstakes cash games, and – being the nerd that I am – by June that year I had already bought my first license of PioSolver.

Since then many different Hold’em solvers appeared in the market – MonkerSolver, SimplePostflop, GTO+, and many others. In recent years, we saw the rise of GTO Wizard, which started as a webservice of PioSolver solutions (therefore, it wasn’t a solver itself) but now also offers their own AI-powered hold’em solver.

During this time, studying with GTO solvers has become the norm amongst professional poker players. Everyone studies with solvers, and a lot of the times, only with solvers. If you browse poker forums, you’ll see microstakes players who make bets of a few cents at a time at the tables discussing GTO concepts they took from the programs they pay hundreds of dollars to use.

Such obsession with these programs has certainly accelerated the improvement in the level of play in online cash games (and I imagine for tournaments and other formats too). For example, I’ve heard from multiple people, and read from multiple sources (including iconic 2+2 threads from legends of the game) that small bets on the flop were almost not a thing in the pre-solver era. That’s crazy to think about because small bets are the most frequently used at equilibrium on flops, in pretty much any postflop spot. For a few years now, even players at the lowest stakes implement small bets in their strategy. Solvers revealed a huge strategical leak in the games of most humans, including the most advanced players at that time, and made that information accessible through a few hundred dollars of paywall plus a few minutes of intense computation.

While solvers have contributed to elevate the overall level of play of most players, it seems like there’s still a lot of room for more. For instance, 10-15% pot bets have only very recently become more popular amongst professionals at the highest levels (these tiny bets had already been predicted by Bill Chen on Mathematics of Poker 10 years before solvers were a thing), even though people had almost 9 years to find out about them by using the solvers. The same can be said about flop overbets. Solvers are here since 2015, but it was only in recent years that some regulars started using them more frequently in their gameplans.

What I sometimes wonder is: how far are we, currently, from computer performance? In these almost 9 years of nash equilibrium strategies being accessible to the community, how close did we actually get to the computers?

That’s the question I want to answer in this post.

To represent the humans in this comparison, I picked the top 15 cash game pros (in terms of dollar earnings) and combined them into one single alias. If you follow poker media, you can probably name several players from this list without much effort.

I got all of their hands played on GG Poker since 2022. I chose GG Poker because it’s the site with the most high stakes action in the world right now; and 2022-now seems like a reasonable time frame to represent the current state of the game while getting a decent enough sample size.

These 15 players together played almost 1.5M hands during this period, and won over 8 million dollars in the process:

If you’re not impressed by the 2bb/100 all-in EV winrate, keep in mind that they probably got around 2-3bb/100 of rakeback from GG Poker, so the total amount won is likely double than what’s in the print. Not bad.

To represent GTO in this comparison, I used the GTO Hand Histories I generated programmatically through my own custom software. In this program, 6 bots played against each other for 100,000 hands, generating a dataset equivalent to 600k hands.

An important thing to note here is that these bots played according to precomputed Jesolver solutions – which I ran for the entire game tree, all spots, all 1755 strategically different boards. Although certainly a good representation of GTO, it’s important to realize that this is not a true representation of what nash equilibrium looks like for poker, simply because any tree you run in a Solver is a limited version of the whole potential equilibrium tree. The solver restricts the amount of betsizings you can include in your solutions, so any solver tree is at best a partial representation of the true equilibrium.

With that out of the way, we can start our comparison. We can divide this problem in 2 categories – frequencies and range compositions.

At the end of the day, this is what poker strategy is. A description of the frequencies of the different strategic options you play in any given spot, and the composition of the ranges you construct for each of those strategic options in those spots. There is nothing more to it. Frequencies and range compositions.

For this post, I’ll focus on the frequencies, which are easier to analyze with data.

FREQUENCIES

For this post, I’ll analyze and compare the frequencies of the PFR SRP IP – BTN vs BB Spot. I had to pick one spot because otherwise this post would get way too long, and I don’t want to bore you with hundreds of numbers. And I picked this one specifically because it should be the spot that professionals perform the best – it’s the most frequently played one, and the one people tend to study the most. I wanted to give humans their best possible chance here.

Let’s start with Flop Cbets:

Here we can already notice several differences between GTO and our best players, which I’ll call the crushers from now on:

The crushers cbet way more than solver:73% x 62%. Almost all of the difference comes from the excessive use of the 30% sizing – 51% for the crushers, while only 42% for the solver;
Likely due to the previous item, the crushers fold way more to check-raise after betting small:43% x 33% (this comes from the FCR numbers in the right-side column);
The crushers are still not using the flop overbets in their gameplans:10% for the solver, only around 2% for the crushers;

Following flop cbets, let’s look at how they compare facing turn probes:

With regards to the raising frequencies, the crushers are performing quite close to equilibrium:

12% raise facing small probe, compared to the same 12% from GTO;
4% raise facing 75% probe, compared to 3% from solver.

When it comes to their folding frequencies, however, signifcant differences are observed:

Facing small probes: 28% x 18%
Facing 75% probes: 45% x 36%
Facing Overbet probes: 63% x 54%

Next, we take a look at the Bs B node, in other words, the double barrel frequency after betting small on the flop and getting called:

In this node, my solver trees had only 3 betsizing options: 70% (referred to as Big Bet), 120% (Small Overbet) and 1.8x.

Here we observe the following differences:

The crushers double barrel more often:46% x 42%;
The crushers fold less to turn check-raise after betting 3/4:46% x 56%

Lastly, let’s take a look at some river stats. First, how they respond facing river probes (OOP goes XC-X-B):

We see differences again in the folding frequencies:

Facing river probe after flop small bet: 59% x 52%;
Facing river probe after flop big bet (75%): 46% x 42%

And also differences in the triple barrel and BXB frequencies:

Triple barrel after betting small flop: 50% x 45%;
Triple barrel after betting big flop: 46% x 60%;
BXB after flop small bet: 39% x 38%;
BXB after flop big bet: 46% x 42%;

I’m going to stop right here so that you don’t get nauseated with so many numbers. But what we saw above is sufficient to observe 2 very clear patterns:

Our human representatives are betting more frequently than solver (higher flop cbet, higher turn cbet, higher river Bs B B, higher BXB);
They’re also folding more often than solver (higher fold to check-raise, higher fold to turn probes, higher fold to river probes).

If you have watched my videos and followed my content on YouTube, this may not be big news to you. A month ago for example, I posted a poll in the community tab essentially asking people how well they thought High Stakes regs defend against small flop cbets when they’re in the BB against SB in a single raised pot:

Only 20% of the people answered correctly – they’re folding 33% on A high low low boards, which constitutes an 11% overfold relative to solver in those textures. In the aggregate of all textures, solver folds 19% to small flop cbets, and the high stakes regs fold 28%.

When I post these numbers, someone always comes and say: “well yeah they are not playing GTO but that could just be them adjusting to the player pool”.

I can tell you whether those deviations that the best players in the world are making relative to solver are actually good adjustments vs their opponents or not. In fact, that’s quite easy to determine: all you gotta do is run the numbers for the player pool and see if they are playing in a way that gets exploited by how the high stakes regs are playing. For example, overfolding vs turn and river probes is a good adjustment against a player pool that probes too strong relative to solver. If the pool is in fact probing too strong, then the high stakes regs are exploiting them and therefore the deviation is good.

I’m not gonna tell you in this post if those adjustments are good or not. What I wanted to do with this week’s post was to show you a piece of truth that can save you a lot of time and effort in your poker career. I wanted to demonstrate to you that perhaps the current obsession of the community with GTO strategies actually does more harm than good.

THE BEST PLAYERS IN THE WORLD ARE NOT PLAYING GTO

This conclusion is amazing in multiple levels. First of all, it demonstrates how poker is hard and how even the best players are still playing exploitable strategies. Remember that, regardless of whether or not they are purposefully making these deviations, by deviating from solver they’re making themselves exploitable. Someone playing against them can take advantage of these numbers to build a counter strategy that makes money off of those deviations. Excessive fold to flop check-raise can be exploited by more bluff check-raises. Excessive double barrel can be exploited with flop slowplays with the nuts. Excessive folding against turn probes can be exploited through more aggressive, bluff heavy leading ranges.

If the best players in the world are exploitable – the ones taking 8 million dollars out of the ecosystem every year – imagine everyone else. Imagine how exploitable that 46 WWSF reg you play everyday against is. Imagine how exploitable that recreational that plays 50% of hands preflop is. Everyone is hugely exploitable in poker – in pretty much every node of the game tree.

It’s been almost 9 years since the first solver came out. And yet, nosebleed crushers can’t mimic it. This gives hope to the pessimistic few that think the abundance of content and tools in the internet will kill poker. I would bet my entire net worth that it won’t – as long as sites can be effective at preventing cheating, which is another story. But poker won’t become unbeatable because people got too good from learning with solvers, other poker software, videos or courses in the internet. It simply won’t – because playing close to equilibrium is very, very hard.

At the same time, this conclusion is a snap in the face of the unspoken but assumed idea nowadays that you must play GTO to make money in poker. The data shows very clearly that the people making the most money in the ecosystem are making big deviations from solver, particularly in the directions of being more aggressive than solver with the betting lead, and being more passive than solver without it.

Again, these 2 patterns of deviation may or may not be good, we are not assigning a label to them in this post. These players could be making lots of money despite these deviations, not because of them.

But the fact is that, if the people making the most money in the game are not playing GTO, then surely playing GTO is not a necessary condition to make money.

What I want you to take out of this post is: Exploitative play is where the money is at.

It’s by continuously exploiting your opponents imbalances that you’ll maximize the amount of money you can make in this game. This is not to say that you should neglect theory and GTO concepts – on the contrary, I think learning GTO makes you a better exploitative player. But learning GTO is just the first step in your journey of becoming the best player you can be. At some point you have to start crafting an exploitative strategy that builds upon your baseline strategy to extract EV from your opponents mistakes. That’s what will make you money.

And now I need to ask you: when was the last time you devoted some study hours to crafting your exploitative strategy?

You don’t have to say it out loud. I know it’s been a while, so I’ll spare you of the embarrassment.

My goal as a content creator is to help you in this journey of becoming a better poker player. So, if I want to achieve that goal, I have to be the one to tell you: cut the GTO obsession. Learn GTO for the purpose of having a baseline strategy, period. GTO is the means, not the end. Then you start doing the work that really matters – your exploitative gameplan.

Are Nosebleed Players GTO?

Poker strategy insights, directly to your inbox.