---
title: 'The Best Baccarat Strategy, Tested on 57 Billion Hands'
source: 'https://youtube.com/watch?v=hkotVAAGv9Q'
video_id: 'hkotVAAGv9Q'
date: 2026-11-16
duration_sec: 1231
channel: 'Dice Lab'
---

# The Best Baccarat Strategy, Tested on 57 Billion Hands

> Source: [The Best Baccarat Strategy, Tested on 57 Billion Hands](https://youtube.com/watch?v=hkotVAAGv9Q)

## Summary

This video presents a comprehensive analysis of baccarat, testing 19 different strategies across 57 billion simulated hands to determine which, if any, can beat the casino. The findings reveal that no betting system or pattern-reading strategy provides a real edge, and even perfect card counting yields minimal profit. The only method that truly beats the game is exploiting a physical flaw in the cards, as demonstrated by Phil Ivey's infamous court case.

### Key Points

- **Baccarat's Dominance in Casinos** [00:01] — Baccarat generates over 80% of casino revenue in Macau and won $1.36 billion on the Las Vegas strip in a single year, making it the most profitable table game.
- **Simulation Setup** [00:43] — The creator built a baccarat table simulation with 8 decks, real shuffled shoes, and all casino rules. 19 different strategies were tested, including Martingale, Fibonacci, card counting, and pattern reading, over 57 billion hands.
- **Basic Rules of Baccarat** [01:23] — Baccarat deals two hands: 'player' and 'banker'. Players bet on either hand or a tie. Cards are valued with Aces as 1, 2-9 as face value, and 10s/face cards as 0. The hand's value is the last digit of the sum. A natural 8 or 9 wins immediately. The banker's draw is determined by a fixed table based on its total and the player's third card.
- **House Edge Breakdown** [03:01] — The banker hand wins 45.86% of the time, player 44.62%, and ties 9.12%. The house edge is 1.06% for banker bets (after 5% commission), 1.24% for player bets, and a massive 14.36% for tie bets.
- **Strategy Test Results: Pattern Readers** [04:13] — Pattern-reading strategies (following streaks, chopping, etc.) performed identically to random betting, losing about $85 per night. Reading the scoreboard made no difference; the only edge came from betting on banker more often.
- **Strategy Test Results: Money Systems** [07:01] — All progressive betting systems (Martingale, Fibonacci, etc.) lost more than flat betting. Martingale lost $117 per night and went bust on 74% of nights. The house edge remains constant; progressions only increase the total amount wagered, leading to larger losses.
- **The Myth of Streaks and Hot Shoes** [08:13] — Analysis of 45 billion hands showed that after any number of consecutive banker wins, the probability of banker winning next remains 50.68%. Hot shoes (first 10 decisions all banker) showed no difference. Streaks are common (69% of shoes have a run of 6+) but are just random variance.
- **Martingale with Unlimited Bankroll** [10:37] — A Martingale bot with $10 million and no table limit won 95% of nights but still lost an average of $662 per night due to rare catastrophic losing streaks. With a $5,000 table limit, it won 77% of nights but lost $357 per night on average.
- **The Kashiwagi-Trump Duel** [12:04] — Akio Kashiwagi's $10 million freeze-out against Trump's casino is analyzed. Simulations show that with small bets, he had only a 7.9% chance of doubling his money, while larger bets increased his chances to 27%. The casino's edge grinds down over many hands.
- **Side Bets and Variations** [14:00] — No-commission baccarat variations have different house edges: EZ Baccarat is 1.02%, but the version paying half on banker 6 is 1.46%. Side bets like Dragon 7 (7.61% edge) and Panda 8 (10.19% edge) are terrible. The best side bet is the Dragon Bonus on the player side at 2.65%.
- **Card Counting in Baccarat** [15:22] — Card counting is nearly useless for the main bets because every card affects banker and player odds almost equally. A perfect card counter made only 4 cents per hour. However, a simple count for the Dragon 7 side bet (counting 4s-7s as -1, 8s-9s as +2) can yield an 8% edge, but it's slow and likely to get you banned.
- **Phil Ivey's Edge Sorting Method** [17:11] — Phil Ivey exploited a flaw in the card design (non-symmetrical pattern) to identify high cards (7, 8, 9) before they were dealt. This gave him a 6.48% edge over the casino. He won £7.7 million in London but lost the court case, and was ordered to repay $10.1 million in Atlantic City.
- **Final Recommendations** [19:28] — The best strategy is to bet on banker with a flat bet, avoid side bets, and look for a 4% commission table (house edge drops to 0.6%). For a specific goal, a few big bets are better than many small ones, as the casino's edge needs time to work.

## Transcript

a single decision. You [music] don't hit. You don't fold. You don't choose a card. You put your money on one of two hands and the cards do everything else. And somehow [music] that game runs the richest casino city on Earth. In Macau,
Bakarat brings in more than eight out of every [music] $10 the casinos make. On the Las Vegas strip, it won $1.36 [music] billion in a single year. That's more than a third of everything the strip's tables won. It's the game James
Bond plays. [music] It's the game where a Japanese real estate tycoon took on Donald Trump's casino for $10 million. [music] And it's the game where Phil Ivy won 7.7 million. And then the highest court in Britain called it cheating.
After my blackjack and poker [music] videos, you kept asking for this one. So I built a bakarat table inside my computer. Eight [music] decks, real shuffled shoes, every rule exactly like a casino. Then I sent in 19 different
strategies. [music] The scorecard readers, the streak chasers, martingale, Fibonacci, a card counter, and one bot that cheats the way Phil Ivy did. [music] 57 billion hands later, I know
which of them works. Spoiler, it's almost none of them, but the reasons why are a lot stranger than [music] you'd think. A Martingale player with $10 million won 95% of his nights and still lost [music] money. and a perfect card
counter, one that knows every single card left in the shoe, [music] made four cents an hour. Let's start with how this game actually works. Bakarat deals two hands. [music] One is called player, the other is called banker. And here's the
first thing that confuses everyone. You're not the player, and the casino isn't the banker. Those are just [music] names. You can bet on either hand or on a tie between them. Each hand gets two cards. [music] Aces count as one. 2
through nine count as their number. Tens [music] and face cards count as zero. Then you add them up and you only keep the last digit. A seven and an eight make 15. So that hand is worth five. The best possible hand is [music] nine. If
either hand has an eight or a nine on the first two cards, that's called a natural and it's over. Nobody draws. Otherwise, the player hand draws a third card on [music] anything from 0 to 5 and stands on six or seven. Then comes the
part almost nobody knows. [music] Whether the banker draws depends on its own total and on the exact card the player just drew. A banker with three draws unless the player's third card was an eight. A banker with six only draws
if the player's third card was a six or [music] a seven. It's a fixed table. The dealer follows it and you never touch [music] it. So why do people sit at these tables for hours? Partly the money and partly the ritual. [music] At the
big tables in Asia, high rollers are allowed to squeeze their cards, peeling up one corner, then one edge, [music] so the number appears a little at a time. It changes nothing. It's still the most dramatic thing in a casino. My table
plays [music] all of it. Eight decks, a burn card at the start, a cut card near the bottom, and every [music] third card rule. Before trusting it, I worked out the exact odds of every possible first [music] hand from an eight deck shoe
card by card and compared them to the simulation. [music] They matched to the fourth decimal place. And this is what they say. The banker hand wins 45.86% [music] of the time. The player hand wins 44.62.
And about 9 12% of hands are a tie. Banker wins more because [music] it plays second. It gets to see the player's third card before it decides. casinos [music] paid banker bets even money, they'd lose. So, they take a 5%
that commission, here's what [music] each bet costs you. Banker 1.06% of every dollar you bet. [music] Player 1.24%.
And the TA, which pays [music] 8:1 and looks so tempting on the felt, 14.36%. That one number is going to matter a lot. Now the players [music] I built 19 of them. Banker only who bets banker every single hand. Player only
who does the opposite. [music] Tai only who chases the big payout. And coin flip [music] who picks banker or player at random to show what no strategy at all looks like. [music] Then the pattern readers follow the shoe
bets on whatever just won. [music] Chop bets on whatever just lost. Streak writer waits for three in a row and jumps on the streak. Gambler waits for four in a row and bets against it [music] because surely it has to end.
results [music] and decides whether the shoe is streaking or chopping. Then the money systems. [music] Martingale doubles after every loss. Peroli lets wins ride up to three in a row. Fibonacci, Dalbear [music]
1326, Leoucher and Oscars grind each raise and lower the bet by their own formula. [music] And finally, three side bet fans. Pair hunter, who bets banker [music] plus a small bet on both pair bets every hand. Tie chaser, who bets
the tie right after a tie. [music] And banker plus tie, who bets banker And banker plus tie, who bets banker with $5 on the tie every hand. Each one [music] starting bet, a $5,000 table maximum,
and 300 hands. And each bot played 12 million knights. Here's round one. at the bottom with almost no surprise. [music] Tie only the average night costs [music] out of a thousand. 71% of nights end
[music] house edge doesn't hurt you slowly. It eats you. Then the side [music] bet fans pair hunter loses $364
pair hunter loses $364 a night. Banker plus tai loses $287 and 170. Notice that none of them bet much on the side. $5 here and there. [music] That little side bet is enough to double, triple, even quadruple what the
night [music] costs. And now the interesting part. The pattern readers follow the shoe lost $85 a night on average. Chop lost $85. Pattern reader
lost [music] $85. And coin flip the bot that doesn't even look at [music] the scoreboard lost $85.50. reading the shoe, [music] following it, fighting it. It made no difference at all. Streak Rider and Gambler did a
little better at about $80, [music] but not because they read anything. When they aren't chasing a pattern, both [music] of them bet banker, so they bet banker more often. That's it. That's the whole edge. Because the best [music] bot
in the whole league was the most boring one, banker only. It lost $79 a night [music] and it went home up on 42% of nights. player only lost 92. [music] Then the money systems, and this is where people get angry in the comments,
every single one of them lost more than flat betting. Fibonacci lost [music] $87 flat betting. Fibonacci lost [music] $87 a night. Oscars' grind 124, Peroli 123, [music] 1326, 133, Labusher
156, [music] Dalbear 160, and Martingale lost $117 [music] a night. But that's not the scary number. The scary number is that 74% of its nights ended with [music] the whole $1,000 gone. It only went home up
on one night in four. Here's why. I added up every dollar [music] each bot pushed across the table and checked how much of it the casino kept. Banker only [music] 1.06%. Martingale 1.06%.
Martingale 1.06%. Labusher Dalbear [music] Peroli 1326 all &gt;&gt; [music] &gt;&gt; Every banker system loses the same slice of every dollar it bets. A betting system doesn't change that slice. It
only changes [music] how many dollars go through the slicer and the progressions push a lot more dollars through so they [music] lose more. That's round one. If [music] really helps the channel and there's a lot more coming. Now, let's
get to the thing every Bakarat player believes in. Walk past any Bakarat table and you'll see a screen above it full of red and blue circles. Some casinos also hand out little paper cards and pencils so players can write down every result.
[music] The screen has its own language, the bead plate, the big road. And when one side wins over and over, the big road draws a long tail down the screen, and players call it a dragon. A study published in 2025 looked at almost 18
million real Bakarat games from more than 6,000 players. Most of them bet with the streak. They see a dragon [music] and they ride it. So, does the shoe remember? I took 45 [music] billion hands and looked at what happens after a
streak. After one banker win, the next hand goes banker [music] 50.68% of the time, not counting ties. After five [music] bankers in a row, 50.68. After 10 bankers in a row, a monster dragon, still 50.68. [music]
The shoe doesn't know. What about a hot shoe? I found every shoe where banker won [music] the first 10 decisions, more than 600,000 of them and checked the rest of the shoe. [music] Banker won 50.69%.
Normal shoes 50.68. [music] No difference. And yet, dragons are everywhere. 69% [music] of shoes have a run of six or more of the same result. Almost one shoe in four has a run of eight. [music] The longest banker streak
in my whole run was 34 hands in a row. The longest player streak, [music] 35, and somewhere in there, 10 ties in a row. That's the trick. Streaks feel [music] like signals because they're so common. But they're just what random
one. Even though banker is the better bet, the player hand still wins more [music] hands than banker in 43% of all shoes. There's one more reason tables feel hot or cold. [music] Picture seven players at one table, all betting
banker. They all get the same cards, so they all [music] win or lose together. In my sim, all seven went home up on 42% of nights. [music] If each of them had played a different shoe, all seven being up would happen on
fewer than one night in 400. [music] A hot table isn't seven lucky people. It's one coin flip that seven people share. So, patterns [music] don't work. But what if you just had more money than the casino could ever beat? [music]
Martingale is simple. Bet 25 on banker, lose and bet 50, lose again, bet 100, [music] keep doubling until you win. And every win gets back everything you lost, plus [music] a profit. With enough money, you should never lose. So, I gave
a Martingale bot $10 million, [music] removed the table maximum, and let it play 12 million nights. It went home up on 95% of [music] them, 19 nights out of 20. It was a genius. And it still lost an average of $662 a night because on
the other night in 20 things [music] can go very wrong. Some shoes produced a losing run long enough to burn [music] through the whole thing. In my sim, the through the whole thing. In my sim, the worst night lost all $10 million in one
evening, chasing [music] a $25 profit. The longest losing run for a banker bet in my whole run was 35 hands in a row. To survive 35 losses [music] in a row starting at $25, the next bet alone would be more than $800 billion. [music]
And real casinos don't let you get that far. With a $5,000 maximum, [music] far. With a $5,000 maximum, [music] $25 doubles up to 3,200. And the next double simply isn't allowed. [music] Then even a $10 million martingale
Then even a $10 million martingale player wins 77% of nights and loses $357 a night on average. Which brings us to the most famous Bakarat [music] duel ever played. In January 1990, Akio Kashiwagi, a Japanese [music] real
estate investor, won about $22 million at a casino in Darwin, Australia. A month later, he walked into Donald [music] Trump's casino in Atlantic City, the Trump Plaza, and left 2 days later, up 6.2 million. Trump wanted [music] him
back. They agreed on a freeze out. Kashiwagi would keep playing until he either doubled his money or lost it all. How much money depends [music] on who's telling the story. Trump said 10 million. Kashiwagi said 12. [music]
Reports of his bet size go from $100,000 a hand all the way to $250,000. [music] So I ran all of them. A $10 million freeze out at 100,000 a hand. [music] Kashiwagi doubles first just 7.9% of the time [music] at 200,000 a
hand with 12 million at stake 18%. Even 250,000 he wins only 27% of the time. [music] And notice the pattern. The bigger the bet, the better his chances. Over a long
match, that 1% edge grinds you down. [music] Over a short one, you're mostly flipping coins. A freeze out with small bets [music] isn't a duel. It's a slow bets [music] isn't a duel. It's a slow execution. In May of 1990, after about 6
days, [music] Kashiwagi lost $10 million. He paid six. Less than two years later, he was found murdered at his home near Mount Fuji. [music] The case was never solved. You can test the same idea with your own $1,000. Say you
want to turn it into 2,000. Bet 25 a hand on banker and you get there [music] only 27% of the time. Put the whole thousand on player. one single [music] thousand on player. one single [music] hand and you double it 49% of the time.
If you absolutely must gamble for a goal, [music] a few big bets beat a lot of small ones. The casino's edge needs time to work and small bets give it all the time in the world. Now, the bets on the edges of the table. Lots of casinos
[music] now offer no commission. Bakar easy Bakar pays banker wins even money but if banker wins with three cards [music] totaling seven the bet just [music] totaling seven the bet just pushes its house edge is 1.02% [music] a
tiny bit better than the normal table but the other popular version where a banker win with six pays only half costs 1.46% 46%. [music] No commission sounds like a gift. Check which one you're sitting at. And the
side bets are where the money disappears. The Dragon 7 pays 40 to1 and [music] costs 7.61%. The Panda 8 pays 25:1 and costs 10.19.
[music] The player pair and banker pair 10.36% each. Either [music] pair 13.71. The only side bet in my test that came close to [music] fair was the dragon bonus on the player side at 2.65%. [music]
At $25 a hand, every 100 hands on banker costs you about $26. Every 100 hands on the tie costs you about [music] 359. And if you can find a table that takes
4% commission instead of five, banker's house edge drops to 6%. [music] That's the single best number in this whole video. But blackjack players beat the casino by counting cards. [music] So why not Bakarat? In the 1960s, Edward
the mathematician who proved blackjack could be beaten, studied Bakarat with William Walden. They found one side bet that counting [music] could beat, and casinos quickly removed it. But the main bets almost never. Every card that
leaves the shoe [music] moves the odds of banker and player in almost the same way. So the two bets barely drift apart. So, I built the best counter [music] possible, not a person. A computer that knows every single card left in the shoe
[music] and calculates the exact odds of every bet before every hand. It only bets when the math is in its favor. On the banker and player bets, that happened on less than half a percent of hands. Its profit [music]
at $25 a bet, 4 cents an hour. The perfect card [music] counter with information no human could ever hold in their head made 4 cents an hour. But then it looked at the side bets and the dragon 7 was different. A banker win
with three cards totaling seven needs lots of small cards, especially fours, [music] fives, sixes, and sevens. When those pile up near the end of the shoe, the dragon seven flips to the player's side. So, I gave [music] my counter a
simple count a human could actually use. Fours through sevens count minus one. [music] Eights and nines count plus two. divide by the decks left and only bet the Dragon 7 when the count reaches [music] four. It works. Each Dragon 7
bet made in those spots is worth over [music] 8%. It's a real edge against the [music] 8%. It's a real edge against the casino. It's also about $15 per shoe at a $25 bet. And a shoe can take more than an hour to play. You're beating the
casino at roughly minimum wage. [music] And the moment they notice you only bet Dragon 7 near the end of the shoe, they'll ask you to leave. So, is there any way to really beat this game? There is one. And it ended up in court. In
August [music] 2012, Phil Ivy, one of the best poker players alive, sat down at a Punto [music] Banko table at Crockford's casino in London with a noticed something [music] about the
cards Crockford's used. The pattern on the back wasn't perfectly symmetrical. One long edge looked slightly different from the other. If a card was turned around, you could tell. So she asked the dealer in Cantonese to turn certain
cards around as they came out, saying [music] they were good cards. Those cards were sevens, eights, and nines. Crockfords agreed to her other requests, too, including that the cards were shuffled by a machine, which doesn't
[music] turn cards around. Shoe after shoe, the turned cards stayed turned. For the first 79 hands, Ivy gained nothing. [music] That was just setting up the deck. After that, they could often tell whether the
first card out of the shoe was a seven, [music] eight, or nine before it was dealt. In one shoe, they called it right 23 times out of 27. [music] So, I gave one of my bots that exact power. It knows one thing, whether the
player's first card [music] is high. If it is, it bets player. If not, it bets banker. That's all. The house edge disappears. The bot gets a 6.48% edge over the casino. [music] The court papers say about 6 12% and my Sim landed
right on it. At $50,000 a hand, every hund hands is worth [music] more than $300,000. And if a bot could read all four opening cards the same way, its edge more than doubles to 13.7%.
Ivy won £7.7 million. [music] Crockford's refused to pay and gave back only his1 million stake. He sued. In 2017, the UK Supreme [music] Court ruled against him. He'd never touched a card, but he'd tricked the dealer into turning
he'd done the same thing in [music] America. At the Borgata in Atlantic City, over four visits in 2012, the same method won $9.6 million. A judge ordered
him to pay [music] back $10.1 million. In 2020, they settled and the terms are secret. [music] So, the only strategy that ever beat Bakar at big money didn't [music] Let's put the whole thing
together. Give My Banker only bought $1,000 in flat $25 bets and it survives $3,719 hands [music] on average. Along the way, it pushes about $93,000 across the table. $93,000 [music] to lose 1,000.
table. $93,000 [music] to lose 1,000. The tie only bought lasts 278 hands. So, here's [music] what 57 billion hands say. Bet banker. Bet the same amount scoreboard because the shoe has no memory. Skip the tie, skip the pairs,
and skip the panda. Look for a 4% [music] commission table if you can find one. And if you want a real chance at a goal, bet big and bet few [music] because time is the casino's best friend. And that's the strangest thing
about Bakarat. It's one of the best bets in the casino. It's also the one with the most systems, [music] scorecards, and rituals. All that thinking around a game where none of it matters. If you enjoyed [music] this, subscribe and tell
me in the comments which game I should test next. I'm reading all of them.
