The game of Bond, of Macau's billion-dollar rooms, and of people squeezing cards like they're reading tea leaves. Underneath the ceremony it's the simplest game in the building: two hands are dealt by a fixed rulebook, you guess which will be closer to nine, and you have exactly zero influence on anything. The edge is low. The decisions are none. The pace is fast.
Two hands are dealt, called Player and Banker — neither is yours. Cards count at face value, tens and pictures are zero, and only the last digit of the total matters (7 + 8 = 15 = 5). Each hand gets two cards; a fixed table of rules decides whether either draws a third. Closest to nine wins. You bet on Player (pays 1:1), Banker (pays 1:1 minus a 5% commission), or Tie (pays 8:1). That's it. There is no strategy because there are no decisions; the "drawing rules" happen automatically.
Banker wins slightly more often than Player (about 45.9% vs 44.6% of all hands, the rest ties) because Banker draws last and reacts to Player's card. The 5% commission exists to claw that back, and it doesn't quite manage it: Banker at 1.06% remains marginally better than Player at 1.24%. That is the entire strategic content of baccarat, and I've just given it to you.
| Bet (8 decks unless noted) | House edge |
|---|---|
| Banker (5% commission) | 1.06% |
| Banker, single deck | 1.01% |
| Banker, EZ / commission-free (3-card 7 pushes) | ~1.02% |
| Player | 1.24% |
| Banker at 4% commission (rare, promotional) | 0.60% |
| Tie at 9:1 | 4.84% |
| Dragon 7 (EZ side bet) | 7.61% |
| Panda 8 (EZ side bet) | 10.19% |
| Player / Banker pair (11:1) | 10.36% |
| Tie at 8:1 | 14.36% |
Pick a bet, a unit and a number of bets. The simulator plays 1,000 sessions from the bet's actual outcome table — not a normal approximation — and shows the spread, the drawdowns, and how long it takes before the house edge stops hiding behind luck. Banker at 1.06% looks like nothing per hand. Run 700 hands — a long evening at the big table — and see what nothing adds up to.
Card counting doesn't work here. The drawing rules are so symmetric that even a perfect count produces a playable edge on Banker or Player almost never; the Tie can be counted, but the opportunities are vanishingly rare. The academic conclusion is that baccarat is effectively count-proof.
Edge sorting is the famous exception. Some card backs are printed with a pattern that's asymmetric by a fraction of a millimetre. If a player can persuade the dealer to rotate the high cards 180° ("for luck") and the casino uses an automatic shuffler that doesn't turn cards, the player can identify high and low first cards from the back before betting — worth a player edge of roughly 6–7%. A world-famous poker player and a partner used it to win about $10 million from an Atlantic City casino and about £7.7 million from a London one in 2012. Both casinos refused to pay or sued to recover. Both won: the UK Supreme Court (2017) and the US Third Circuit (2019) held that the technique was cheating or a breach of the gaming contract, even though the player never touched the cards. The law's line: exploiting the casino's mistake in your head is fine; getting the casino to alter the game for you is not.
Loss rebates are the legal cousin. High rollers negotiate a discount on losses (say, 20% back on a losing trip). One Atlantic City player in 2011 negotiated a rebate and rule set so generous that, combined with a disciplined stop-loss, he had a positive expectation — and took about $15 million across three casinos, in blackjack rather than baccarat, before they stopped offering. Legal, contractual, and unavailable to anyone betting less than five figures a hand.