Over the last couple of years, AI has been taking over the world. We’ve seen it everywhere. But exactly how good are these AIs at playing the game of poker?
Ten different LLMs have been selected to duke it out in this arena. A lot of these AIs you’ve probably heard of:: Grok-4, Grok-4-fast o3, GPT-5, GPR-5-mini, Gemini-3 Pro, Gemini 3-Flash, Deepseek 3.2, Claude Opus 4.5, and Claude Sonnet 4.5.
These AIs have all played a round-robin match versus each other, and then we’ve seeded them based on who the top eight are, in random order. Their head-to-head match results will determine who makes it all the way to the finals, and ultimately is the champion.
These matches are 20,000 hands each, and they feature mirrored hands, meaning the AI plays both sides of the same hand. So, in theory, it should be a pretty good way to determine which AI reigns supreme.
Now, why should you watch this?
First off, it’s pretty interesting to see what these things can do. We also can see their logic and why they made those decisions, and some of these really are a doozy.
And then also, if you want to just learn poker and learn what you should or shouldn’t do in heads up, or how a top heads up player might break down some of these bots’ strengths and weaknesses, and how I would approach it, I’ll give my thoughts, of course, as to what I think they should be doing in hands, and the way you could approach it if you maybe played a human opponent that played similar.
Breaking down equilibrium strategies and how to exploit people using their statistical tendencies. This isn’t nerdy at all.
We covered a previous matchup between the LLMs that you can browse:
DeepSeek vs. o3: Blocks ?
First Hand: o3 raises JTo, standard, DeepSeek 3-bets KJs, also standard. o3 calls.
- Flop :
Both players have overcards. Obviously, o3 with the gutter. So the c-bet from DeepSeek on the flop is reasonable. Generally speaking, you’d probably want to mainly check this, but every now and then seems fine. And then o3 makes the call. I think that’s definitely standard here with the overs and gutter.
- Turn:
We see king-jack now go for the barrel, and this is kind of interesting.
I don’t love this. Generally speaking, you want to have some better removal to bet this turn. We’re really only blocking folds with king-jack, right? Because jack-ten, jack-nine, are a little more likely to fold the turn now. And we don’t block flush draws, we don’t really. This is, while it’s a good card for the big blind, I don’t love this turn bet, but we see a bet nonetheless.
And now, o3 just decides to jam all in on the turn here, repping, I guess, the . I don’t understand this. You don’t really have many deuces preflop, so you’re kind of just representing a bluff.
And let’s see what o3 thinks about it.
So the logic is that we’ve seen some loose three-bets: king-three off, eight-three off, ten-three off.
It’s funny. This reminds me of bad tournament players when they say, “Oh, this is the guy that showed down queen-seven in that hand, so of course I had to jam.” Not to mention eight-three off would be a pretty good hand here, but regardless.
Also, we have some interesting logic that the paired two reduces his overpair advantage
Also, we have some interesting logic that the paired two reduces his overpair advantage. Don’t really understand what that means, and makes our strong made hands more believable. So what feels like it’s happening here is we’re looking at this from almost like a full ring mentality, right?
Because in heads up, it’s very, very hard to have hands like full houses, particularly on boards like this. Like you don’t really have and all that often. And those hands actually, once in a blue moon, will four-bet as well, although they do primarily call.
So really, when you jam here, you’re kind of saying you have like a or some kind of draw, or maybe like . You’re not really saying you have boats, because if I have a boat, I want to just call and let him bluff and/or realize his equity.
Our hand – an open-ended straight draw (8 outs ~ 17% equity) + two overcards to his bluffs. Zero showdown value if we don't hit → a perfect candidate for a semi-bluff. Pot odds: call 36 to win 144 → we need 25%, and we only have 17%, and we can't rely on realization against river barrels. The call is marginal; we need to be able to successfully draw or bluff on the river to put him in the black.
So, our hand: jack-ten. Open-ended straight draw. Not an open-ender. It’s a gutter. So, looks like our jobs are safe, boys. The AI can’t read the flop or the turn. Zero showdown value if we miss and perfect semi-bluff candidate.
And exploitative factors: the villain folds quickly to turn/river raises in previous hands (#38, 43, 44, 28, 32). He'll only continue with strong hands, but even against them, our draw has 25-30% equity.
So something worth noting, the AIs are trying to play optimal, but can exploit each other. So, we will see some adjustments being made, but making adjustments off of really small hand samples is not usually something that you want to do, because it leads you down this road towards just punting pretty recklessly.
Okay, the rest of it doesn’t really seem too interesting.
The opponent's range after the turn bet is still very bluff-like due to the wide 3-bet and small postflop sizing. Confidence: Medium.
Oh wow! Average, then.
Our equity in case of a call is ~25%. Confidence: high.
Everything is wrong.
Okay, so. But, after all, the villain still has king-high. The bluff might work. It would seem... But no, there's no chance – DS calls the all-in with a king high. And we're all-in and called for 100bb on the turn, where neither player has at least a pair. Let's look at DeepSeek's reasoning.
Opponent profile (based on 46 hands): Player 0 appears moderately aggressive, capable of bluffing the turn and river (for example, in hand #20). He doesn't overly tighten up postflop, but often gives in to aggression after missing the board.
That’s all just wrong.
My range: I 3-bet from the BB, so my range is strong, but includes a lot of overcards like On the flop, I made a small (half-pot) continuation bet, which is standard when my range has an advantage (overpairs, sets, overcards, some flush draws). The turn is a blank, which increases my edge. I continue betting for value with overpairs, sets, and—as a bluff—overcards.
Villain's Range: He called a 3-bet from the small blind, so his range includes pocket pairs (maybe – , Sometimes – ), suited connectors (like 9Ts and T9s, sometimes with clubs), rare suited aces, and off-suit broadways. After calling on the flop, he's more likely to have pocket pairs ( – ), some (maybe A8s, 98s), flush draw (clubs), rare And (sets) and floats with overcards. Given the dry texture, going all-in on the turn is very strong.
In theory, all-in should be balanced, but oh well.
Possible value hands are all of the boats and ace-deuce, if it’s in range, or occasional overpair. However, the shove size is huge, which is polarizing. Either represents a monster or pure bluff with draws that want to maximize fold equity.
This is actually much more logical in the words it’s saying. The board isn’t very dry.
Key points:
The opponent is capable of bluffing with draws or overcards, given the examples of similar aggressive raises in previous hands (#20, 27, 28). Confidence: Medium.
The board is dry, the value range is mostly sets, and two-pair combinations are few. Confidence: high.
My hand's equity against his bluffs is pretty good, but it's almost zero against value. Against his range , , , , , , , 9Ts my equity is ~30%. If you add some overpairs there ( – ), it will increase slightly.
Overall: This is a tough spot. From a GTO perspective, I should be calling with overpairs and top pairs, and KJs is near the bottom of my continuing range. However, given my opponent's bluffing tendencies and the fact that I'm blocking some flush draws ( blocks combinations ), I'm inclined to call if I believe he's bluffing enough. The pot odds are in my favor if there are bluffs there.
What? The jack of spades blocks the jack of clubs. I don’t even know what that means.
Confidence: Medium: My opponent is bluffing in this spot. High: My equity is close to the required 35%.
So all in all, I don’t hate the analysis. I think it’s decent until the jack of spades thing. But in reality, God, maybe calling is fine. [laughter] I don’t know.
So, as a human, when I look at this spot, I’m thinking to myself: when I bet the turn, they jam. He’s saying he has like ace-eight or some sort of deuce. He’s never going to have boats really, because those just call turn. So I’m just thinking how often is it a bluff versus value. And if it’s at like an equilibrium point, I want to unblock the bluffs. So you really don’t want to have like a , , . You have none of those cards. That’s kind of reasonable.
But you would still just fold because it’s way better just to have an eight here, right? Like you could have , , or , where if you actually are up against it, you actually have outs. And then his pair outs are way worse.
So it’s not the craziest call ever, but it’s still pretty punty from everyone here.
DeepSeek vs. o3 – Three Diamonds = Flush Draw
A couple more hands with thoughts from the AI. DeepSeek raises 6 with , o3 raises it up to 20 , and DeepSeek calls. So far so good. We usually don't want to 3-bet A7o, but it's not a disaster.
- Flop (40):
Continuation bet 12, call, float – everything is good.
- Turn (64):
Second barrel, 44, and raise, 110?!? Call.
- River (284):
Check. All-in 58. Call.
The river actually does make some sense here because you have so much in there. Why not, I guess? But this turn bet call, the turn bet is whatever. I don’t particularly love it, but I don’t hate it.
We can just kind of move past that. It’s not really that important either way.
But the calling off over half of the stack, leaving 58 chips behind with ace high. Let’s see what it thinks.
Opponent profile: Loose-aggressive, capable of semi-bluffing. Folds to river all-ins, but plays tenaciously with draws. Not a straightforward nit.
Wow, the AI is insulting his opponent now. Yeah, this guy is not a straightforward nit. Actually, I guess it’s kind of a compliment really.
Range intersection: opponent raise = sets ( , , ), rare slowplays with / , some ( , ), rare K9s. ~15-20 combinations. Bluffs/semi-bluffs: heart draws, diamond draws, combo draws. ~14-18 combinations (our ace of diamonds blocks his strongest diamond draws). Overall, bluffs and semi-bluffs make up a significant portion (40-50%) of the turn raising range.
Equity vs. Sets and Strong Hands ~22%, against draws ~60-65%. Weighted equity ~41-45%, comfortably above the required 37% (66 for a pot of 177).
An all-in on the turn would require our opponent to call just 58 with 25% pot odds, which he'll easily do with value and most draws, so our all-in lowers the expectation. Folding burns the equity of the nut flush draw.
Plan: Call 66. SPR on the river is ~0.24 (58 in 243). If we make the nut flush, we'll shove; with an offsuit ace, we'll often shove or call. We'll think about blanks, but most often we'll check-pass, and sometimes bluff against good blockers.
So, I think maybe it thinks we have a flush draw. Okay, well, that one is an interesting turn call.
DeepSeek vs. o3
I want to preface this hand with: it might end up being my favorite hand from the entire challenge. It’s going to be kind of hard to top this one, but this is really the peak of AI. This is like, if you thought, man, automation is coming, you got to check this hand out because it is a doozy.
Raise from o3, call from DeepSeek, both flop well.
DeepSeek leads here. This is just not good.
You never want to have leads in this spot. So I don’t really understand this strategy. But regardless, now o3 raises. This is just a good raise, right? You knock out hands that have you beat with showdown value. You have a ton of equity. This is just a very good play here with the gutter and flush draw.
- Turn (50):
DS checks. o3 barrels 35. I like both the bet and its size. So far, o3 is playing the hand very decently. But then the train starts to go off the rails.
We now have a check-raise to 50 big blinds. This is yet another hand where I see the AI just leaving weird chips behind.
In theory, if you look at solutions, you don’t really do raises that leave like some taxi cab fare home. You either call, or you raise small, or you go all in. Because this obviously commits you with a variety of hands. So you’re basically doing the same as just jamming yourself. You basically just don’t do this. This is effectively a jam but with some of the stack behind.
The hand that they choose, I think, is actually fine. You block , , . So I don’t think the logic there is too interesting. But now we have a kind of weird spot for , right? You’re getting roughly two-to-one in position. You have the flush draw and the gutter. You probably have to call. It’s ugly.
You’re not happy about it, but I think you do end up having to make the call.
And this is just a wonderful street of poker. This is again AI at the absolute pinnacle, cutting-edge technology here, because we have an all-in and a call.
o3 calls this all-in with seven high on .
We've got to take a look at the breakdown here, what exactly they were thinking, because I just need to understand this genius play.
Decision: Call. Villain regularly turns missed draws into bluffs on the river. His diamond draws missed. I need to call 75 for 360, meaning a little over 20% of his bluffs are sufficient. Given his LAG style, weak turn minraise, and small river shove, I'm confident that missed diamond draws alone are enough to make my call profitable. None of his value hands will take a line with such small sizings, so the math and villain's tendencies make calling clearly profitable.
What’s truly wonderful about this logic is that we have the three of diamonds. So the only diamond hand that you possibly beat is , which mainly three-bets preflop. So there kind of aren’t any diamond hands you beat. In fact, if your opponent is bluffing with missed diamonds, you’re probably behind.
“I’m convinced this missed diamond portion of their range alone clears that threshold.” None of their value hands are likely to take this line for such a low price. So the math plus their tendencies make calling clearly plus-EV, guys. It’s a clear plus-EV call.
What was DeepSeek thinking going all in here?
I'm going all-in. Given my small remaining stack, I can put a lot of pressure on one-pair hands, and my aggressive check-raise on the turn should add credibility to my story. My opponent has been folding a lot on the river, so I think going all-in will force him to fold all but the strongest hands.
Just a beautiful, the Mona Lisa of poker here, guys.
Grok4 vs. Gemini 3 Flash
Coming into this one, I think this looked a little more reasonable than the last one. I’m going to avoid the statistical analysis just because we’re taking too long per match.
But basically, we have our next quarterfinal match here.
Now, if you’re not familiar with heads-up play, when you have the worst possible hand all-in equity-wise, that could just be a fold to a three-x. That’s my professional opinion. I’m glad they brought in the expert for that level of analysis.
Gemini calls.
- Flop (36):
Grok bets 12 with nothing, calls.
- Turn (60):
The hand went both ways, Grok bets 40. Call.
- River (140):
I actually want to see preflop. Why did Grok three-bet this?
Current situation: We're playing heads-up no-limit Texas Hold'em with an effective stack of 100bb (200 chips, blinds 1/2). I'm in the big blind with a very weak offsuit hand with poor postflop playability. My opponent in the small blind (button) raises to 6bb (3bb, a typical HU open). There are 8 chips in the pot, I call. My choices are fold, call 4, or raise (minimum 10, but strategically higher).
Sounds like a pass so far!
Opponent Profile: Based on 15 hands played, my opponent plays loose-passive preflop from the small blind, raising with a wide range (e.g., 65s, A2o, KTo, J5o), but often folds to aggression. He raised four times from the small blind, I folded twice, called once, and three-bet once. Postflop, he plays tenaciously with made hands but folds to extended aggression. He appears straightforward, not overly aggressive, and potentially vulnerable to light three-bets due to his high fold equity.
You know what I find really interesting about this, guys? Just to take a pause here for a moment.
You’d think AI would be super data-driven, right? So when I’m looking at someone, think about that totally non-nerdy heads-up display I just showed you with the bajillion numbers. The way I always played heads-up coming up, before we had equilibrium strategies available to learn from, I would look at all their frequencies everywhere and try to map a strategy based on their statistical tendencies, because at the end of the day, it’s a math game.
And what’s kind of weird to me about this is they’re not really using math when it comes to their opponent. They’re using human-like read tendencies, like, "I saw jack-five, I saw king-ten, he folds to pressure."
The opponent opens wide from the small blind (~40-50% of hands, judging by our history).
This isn't broad at all! On the contrary, it's very tight. I expected to see reasoning along the lines of "I need x% equity, and his fold frequency in this spot is y%"—that's exactly how I reasoned during the game. But they're replacing math with word salad.
The opponent has no range advantage...
Hey, he raised preflop! No, this whole thing makes absolutely no sense.
The GTO baseline is to fold about 35% of hands, including off-suit trash like 23o, call 50%, and 3-bet 15%. However, I'm going to exploit my opponent's overfolding tendency and lean toward bluffing 3-betting 32o. Sizing to 18—three times his open—polarizes my range, puts pressure on his marginal hands, and doesn't give me good pot odds. If he calls, I can bluff on textures that are good for me, aiming to force him to fold. A call would be -EV.
That’s wrong. I don’t know how it’s getting the wrong GTO baseline, but that is wrong.
So there is a thing actually, this is a fun story. I played versus Bryn Kenney like 10 or 12 years ago or something. And at that time, min-raising 100 percent of the button had started to fall off in popularity, but there were still people that were doing it. And Bryn Kenney was trying to min-raise 100 percent versus me. And I was just laughing, because you just can’t do that.
Basically, you just defend like 90 percent, and then he folded like 72 percent to three-bets. So I didn’t need equity to three-bet. So I three-bet him like 40 percent, and I only did it with bad hands. And there was a hand he min-raised and I called ace-queen and it showed down. And I just sheepishly looked around and was like, “Hope nobody was watching that,” because if you see someone three-betting 40 percent and they call ace-queen… Well, you get the idea.
Um anyway, so that does exist in theory. It exists, but it’s very rare.
Anyway, okay. So basically, they’re going to try and take advantage of their folding tendencies. And then the river: opponent appears recreational. I don’t think that the AI is recreational. Opens wide from the button. Calls the bet somewhat loosely. Wait, that contradicts the earlier thinking. Hold on. This is a direct contradiction. Calls three-bet somewhat loosely and then preflop high fold equity. Those are opposites.
It kind of feels like Grok just wanted to go to town, so every street is just like yeah, and then uh, and then uh, I’ll get the fold here. But yeah, river bet is not good either. When you have pairs on brick boards like this, you basically never jam river because you have showdown value. Like if Grok check-called here, I could be like, all right, I kind of see it, right. The straights don’t matter, but it’s always good to have the middle blocker to that.
But you don’t have a non-paired heart. You don’t have a club. You don’t have a or a . You don’t have a or a . I actually do not mind this as a bet, bet, check-call. But as a bet, bet, I hate this. I hate it so much.
And the answer is no. Gemini Flash does not fold much to three-bet. He opens 67 and folds at 36. So he does not fold to three-bet. That three-deuce is a serious, serious, serious punt.
Grok three-bet 52 percent of the time and then barreled off 87, 70, 58. That makes sense with . I mean, this is like, this makes no sense. Gemini opened tight into a complete absolute maniac and then didn’t fold the three-bet. So great counter from Gemini. Although they only four-bet 6 percent, which seems bad, but man, I kind of like how Gemini Flash played this.
But let’s check out the results.
And this was a really well-played game from Gemini 3 Flash. I believe this is the biggest win rate of any AI versus any other AI in the first round. And you really don’t have to wonder why. He basically just played tight and then had a complete loose maniac go absolutely ballistic and then just stacked him repeatedly, which, you know, usually in heads up the looser, more aggressive player wins. But there is a point where they are just spewing and all you got to do is get your net out and catch them.
Gemini Flash advances the next round with a revolutionary strategy of playing good hands.
Gemini 3 Pro vs. GPT-5.2
This brings us to our next match of Gemini 3 Pro Preview versus GPT 5.2. And I’m going to kick this off with a warning. Okay, I got to warn you. This was the most boring match by a mile. By an absolute mile.
If you thought Gemini 3 Flash Preview played solid standard poker, Gemini 3 Pro Preview takes that to another level. I went through every hand they ended up all in. I think Gemini 3 Pro Preview lost like only three out of 40 or something like that. Five out of 40. It was a very, very, very low number.
Basically, when Gemini 3 Pro Preview puts the money in, he’s got the nuts. Think of Gemini 3 Pro Preview like old man coffee: gets his newspaper, gets a little latte, heads down to the card room, waits for aces. That’s what happened here. And when someone does that, it’s hard to find crazy, wild, insane hands that will blow your mind, right? And that’s the kind of stuff we’re trying to watch.
Here's a typical hand. Gemini opens with a min-raise. Can never be too careful, guys. You only can lose what you put into the pot. Min raise. Obviously, you don’t want to do that.
- Flop (28):
GPT-5 checks, Gemini bets with the gutter. This is a very interesting check-call. Let’s go ahead and just see what the thinking is.
Check-calling king high here. Their small stab after I check in a three-pot should be pretty wide and king-five has enough equity to continue. Mainly the double gutshot.
I see no gutshots. The only gutshot you’re going to get with king-five is the feeling of the money leaving your bank account.
- Turn (48):
A check and a 35 bet from Gemini. Here's what he writes about this decision:
This might be the most legible analysis I’ve seen so far in the match.
Maybe that’s why we’ve seen Gemini 3 play so snug in a lot of spots, because it’s only using very concrete reasons to do things, whereas some of these other ones are just getting way out there. They’re willing to call jams with seven high.
But anyway, we see the barrel here and then GPT, you know, GPT is taking it to the streets, and the streets say when you turn open-ender, you got to go all in. I mean, at that point, pretty much your hands are tied.
GPT says checker is all in. Calling 35 is just too thin. You’re calling in a way that’s thin. I’m only sitting on an open-end straight draw with 17 percent river equity and the pot odds ask for 30. I don’t expect great implied odds because a four or nine is obvious and can kill action.
Jamming is better because their big turn size looks polar and should contain hands that hate stacking off here. One pair, overpairs, and some semibluffs. That’s all fair. And my line range can credibly represent strong value on this texture while denying equity.
And then back over to Gemini 3 Pro Preview. Let’s see what the AI thinks here about calling all in.
I’m calling instantly because I hold the nut straight and cannot be beaten right now. Regardless of what they’re shoving with, I have to get all the money in with the best possible hand.
They only got all in 37 times over 20,000 hands. That is a low, low, low, low, low, low, low number. If we go back to the o3 match, they got all in 343 times. So the o3 match, they got all in almost 10 times the amount of the Gemini Pro Preview match. The Gemini 3 Flash Preview match, they got all in 290 times.
So this is crazy amounts of not getting all in.
And if we look at the hands, first off, 84 percent of those hands, right, and I think there were some chops. Only six losses in the entire sample of getting all in at the end of it. It’s kind of like when I played Negreanu.
When this thing goes all-in, it most often has the nuts. Judge for yourself:
HUD:
And if we briefly take a look at their stats before we know the results: Gemini opened 52 percent, only half of all hands. That is way too tight. Just terrible. And then three-bet 5 percent. Four-bet 3 percent. What in God’s name are we doing?
Meanwhile, GPT was in full aggro mode, opening 100, defending 97, three-betting 41, and then not four-betting, which is smart. You don’t want to mess with that 5 percent three-bet range. That’s probably a little bit too speculative. So in terms of countering here, the 100 percent open is gold versus the guy not three-betting. So that’s amazing from GPT.
But why would you defend 97 versus a 52 percent open? I guess he was min-raising, it seems like. So you’re getting good odds, so this number should be high. But three-betting somebody 41 percent when they’re opening 52, I actually don’t mind this opening strategy if they’re just going to play like that.
The only thing is you have to four-bet more than three percent. That is absolutely ridiculous.
What did he have when he four-bet?
He only four-bet 3 percent, but it was a lot of bluffs. What? That was not what I expected to see. Holy. Lots of , , , ... Wow. Why would you not four-bet the good hands?
Well, now to the results:
Okay, let’s take a look at the results. And the winner is GPT 5.2. Not too hard to imagine when you win 79 percent of hands that you’re going to win the match. And even though the big blind, I think, was really punty the way that this AI approached it, this is just a simple math game. When you win 79 percent of hands, it’s hard to lose.
Lord knows the red line on this one, guys. Look at that red line. Pure red line. Just pure unfiltered red line.
Claude Opus vs. Claude Sonnet
Our final matchup feels a bit like a mirror matchup from Starcraft because we have Claude versus Claude. Opus 4.5 versus Sonnet 4.5. Duking it out. One Claude will remain.
And this is the hand that I liked from this matchup.
Opus raises with , Sonnet 3-bets with , and gets called.
- Flop (32):
Bet with just the . A little loose for this bigger size bet, but that’s fine. Float is fine.
- Turn (72):
Barrel is fine, I think, with the . And then with a nut club, you definitely want to call.
- River (168):
And then this is where it gets pretty interesting. We see an all-in jam and a call, an all-in call, ace-high call down.
Let's look first at Opus's thoughts.
I have the nut flush with the ace of clubs. The only hand that beats me is the exact combination of ten-seven of clubs for a straight flush, which is virtually impossible given that three-bet preflop, easy call.
Can AIs understand flushes?
Four clubs is not a flush. Might have to go back to the drawing board here and make sure that they know that five cards are in a hand. I mean, this is just not true.
Okay, man. So much to unpack there. First off, ten-seven suited is a standard pure three-bet. You can low-frequency call, but it is definitely a standard three-bet. Putting that to the side, it doesn’t know what hand it has. That’s a pretty big issue.
In general, not knowing your combination is a pretty serious problem.
Looking at all big hands that were played between these two AIs, we see only 66 hands got all in, which indicates to me that we have a relatively tight match compared to the other two crazy ones that we had. This seems a bit more balanced, and the ace-queen call-down is pretty wild. But outside of that, most of these look relatively standard.
Let's open the HUD.
And if we look at the HUD for this one, Claude Sonnet opened 92 and Opus opened 88. Two of the more reasonable open stats I’ve seen so far in the entire thing. And they both defended pretty reasonably, 86 and 81, a little loose unless they were min-raising. Yeah, they were min-raising.
Opus won.
Wow, this was actually reasonable, fair-farm-grown, ethical poker. They opened a reasonable amount. They defended a reasonable amount. Three-bets a little bit low, 17 for Sonnet and 12 for Opus. Obviously, you want to be a little bit closer to that 20 to 24 range. Fold to three-bets 70 and 63. Four-bets were kind of low at three and four percent each.
Here you have the final results of all of these AIs. We see the people advancing to the next round: Gemini 3 Flash Preview, o3, GPT 5.2, and Opus 4.5. So many different strategies with opening and how much we’re folding to steals and three-betting.
Quarter-final results:
