Baccarat Odds & Commission

The odds behind banker, player and tie, and how the commission decides which bet is best.

This prices every bet printed on a standard mini-baccarat layout against each other on one scale: the three main bets, the two common commission-free rules, and the side bets. It cannot tell you what any individual shoe will do, and it does not cover house-specific paytables that differ from the ones named here.

Banker commission5%

0%2.69%5.39%10%

Player advantageBanker the better betPlayer the better bet

Banker house edge

1.06%

still the best bet on the table

Break-even commission

2.69%

Player overtakes banker at

5.39%

The gold outline is the cheapest bet on the layout at this commission.

Every bet on the layout, cheapest first

Banker1.06%
Player1.24%
Dragon Bonus, playerside bet2.65%
Dragon 7, 40:1side bet7.61%
Dragon Bonus, bankerside bet9.37%
Panda 8, 25:1side bet10.19%
Either pair, 11:1side bet10.36%
Tie 8:114.36%

Drop the commission, and the banker bet costs

5% commission1.06%

EZ, three-card 7 pushes1.02%

Super 6, a six pays half1.46%

Commission-free is not one game. Pushing the banker’s three-card 7, the EZ Baccarat rule, is the cheapest banker bet of the three, by about four cents per $100 wagered. Halving a winning banker 6, sold as Super 6 or Punto Banco 2000, costs about half a percentage point more than paying the 5%. Note where the first one gets it back: the Dragon 7 side bet pays 40 to 1 on exactly the hand that pushed, and at 7.61% it costs about seven times what the standard banker bet itself costs.

The Dragon Bonus, one schedule on two hands

How often each paying margin of the Dragon Bonus lands, on the player hand and on the banker hand, eight decks
Margin	Player hand	Banker hand
Win by 9pays 30 to 1	0.37%	0.31%
Win by 8pays 10 to 1	0.68%	0.57%
Win by 7pays 6 to 1	1.79%	1.59%
Win by 6pays 4 to 1	2.83%	2.38%
Win by 5pays 2 to 1	3.32%	3.15%
Win by 4pays 1 to 1	3.74%	4.02%

On the player hand

2.65%

house edge

On the banker hand

9.37%

house edge

The Dragon Bonus is worth reading twice, because the two halves of it are not one bet. Backing the player hand costs 2.65%, cheaper than the tie by a factor of about five and a half and the only side bet here in the same neighborhood as the main game. The identical schedule on the banker hand costs 9.37%, about three and a half times as much. The schedule pays for the margin of victory, and the banker’s drawing rules produce fewer wide wins and more narrow ones: every winning margin from five up is rarer on the banker side, while win-by-4, the row that pays least, is commoner. So the good half of this bet sits on the hand the rest of this page tells you not to back, and the bad half sits on the one it does.

Fixed, whatever the commission

Banker wins

45.86%

Player wins

44.62%

Tie

9.52%

Tie 9:1 house edge

4.84%

Read the shoe, can you?

Best hit rate 0%

20 coups. Look at the run of past results and call the next one. Ties are dealt and neither call takes them, so the score is out of coups dealt and the long-run ceiling on it is 45.86%, not 50%. The shoe has no memory either way, see for yourself.

Written and reviewed by Alexander Korsager · last reviewed October 7, 2026 · corrections

The idea

In an eight-deck game the banker hand wins about 45.86% of coups, the player hand about 44.62%, and the rest are ties at roughly 9.52%. Ignoring ties, the banker wins about 50.68% of decided hands, a real, structural advantage that comes from acting last and drawing on better information.

Because the banker bet is favored, the casino claws the advantage back with a commission, usually 5% of winning banker bets.That drops the banker house edge to about 1.06%, still the best bet on the table, a fraction better than player at 1.24%, and far better than tie, which pays 8 to 1 against a fair price near 9.5 to 1 and carries a 14.36% edge.
The commission is the lever.Below about 2.69% commission the banker bet is actually positive expectation; the standard 5% pushes it negative but keeps it ahead of player; and above roughly 5.39% the player bet becomes the better of the two.
Commission-free baccarat is where that lever gets interesting, because it is not one game and the two common versions land on opposite sides of the 5% table.EZ Baccarat drops the commission and instead pushes a banker win with a three-card total of 7, an event that happens on 2.2534% of coups, and that comes to a 1.02% banker edge: cheaper than paying the commission, by about four cents per $100 wagered. The version sold as Super 6 or Punto Banco 2000 halves the payout on a winning banker 6 instead, a far more common hand at 5.3864%, and that comes to 1.46%, about half a percentage point worse than the game it replaced. Neither figure is a rounding difference between house styles; they are two different prices for what looks on the felt like the same courtesy.
It is worth seeing what EZ Baccarat charges for the four cents it hands back.The Dragon 7 side bet on the same layout pays 40 to 1 on precisely the hand that pushed the banker bet, and 40 to 1 against a 2.2534% shot is a 7.61% house edge, about seven times what the standard banker bet itself costs. The rule that makes the base game marginally better exists to create the bet that is far worse, and a player who takes the second gives back the first many times over. On both bases the base game still wins on price: ties push and cost nothing, so per coup actually settled the standard banker bet costs 1.17% and the player bet costs 1.36%, while EZ Baccarat, charged for its extra push as well as the tie, still comes in at 1.15%.
The felt in front of you carries more than three bets, and the published analysis of the rest of it is scattered across four separate pages, which is why almost nobody compares them on one scale. Ranked by cost per unit staked, the whole layout runs: banker at 1.06%, player at 1.24%, then the player Dragon Bonus at 2.65%, then a long gap to the Dragon 7 at 7.61%, the banker Dragon Bonus at 9.37%, the Panda 8 at 10.19%, an 11-to-1 pair bet at 10.36%, and the tie at 14.36%.
Two things in that ranking are worth stopping on.The first is that a side bet can be reasonable: the player Dragon Bonus is cheaper than the tie by a factor of about five and a half and sits closer to the main game than to the propositions it shares the layout with. The second is that the same bet, on the other hand, is not. The banker Dragon Bonus pays an identical schedule and costs 9.37%, about three and a half times as much, because the schedule pays for the margin of victory and the banker's drawing rules produce fewer wide wins and more narrow ones. Every winning margin from five up is rarer on the banker side, while win-by-4, the row that pays least, is commoner. The good half of the bet is on the hand the rest of this page tells you not to back.
None of which makes any of it cheap in the way an hour at the table feels.Baccarat deals fast, about 70 coups an hour at a mini-baccarat table by the pace this site uses across its house-edge table, so a $25 flat bet is $1,750 through the shoe every hour. At 1.06% the banker bet is an expected $18.51 an hour, the player bet $21.61, and the tie, at 14.36%, is $251.29. Those are averages over a long run and not a forecast of any one session, which is the point: the banker bet being the best on the table is a statement about the ranking, not about the cost.

FAQ

Questions players ask

Is banker or player the better bet?

Banker, on a standard 5% commission game: a 1.06% house edge versus 1.24% for player. The gap is small but it is consistent, so banker is the default advice.

Should I ever bet the tie?

No. At the usual 8-to-1 payout the tie carries a house edge around 14.36%, more than ten times banker or player. A 9-to-1 tie is far less bad at about 4.84% but still clearly worse.

Is commission-free baccarat a better deal?

It depends entirely on which rule replaces the commission, and the two common ones land on opposite sides of the 5% game. EZ Baccarat pushes a banker win with a three-card total of 7, which happens on 2.2534% of coups, and that works out to a 1.02% banker edge, slightly cheaper than paying the commission. The version that halves a winning banker 6, sold as Super 6 or Punto Banco 2000, is charging you on a hand that comes up on 5.3864% of coups, and that works out to 1.46%. Read the layout for which rule you are getting, and treat the Dragon 7 side bet the EZ table offers as a separate decision: at 40 to 1 on that same 2.2534% event it carries a 7.61% house edge.

Are the side bets on the layout worth playing?

Mostly no, with one real exception, and the exception is the interesting part. Ranked by house edge per unit staked in an eight-deck game: the player Dragon Bonus is 2.65%, the Dragon 7 at 40 to 1 is 7.61%, the banker Dragon Bonus is 9.37%, the Panda 8 at 25 to 1 is 10.19%, and a pair bet at 11 to 1 is 10.36%. So the player Dragon Bonus is the only one that lands anywhere near the main game, and it is cheaper than the tie by a factor of about five and a half. Note that its mirror image is not: the banker Dragon Bonus pays the same schedule and costs about three and a half times as much, because that schedule pays for the margin of victory and the banker's drawing rules produce fewer wide wins. Cheaper than the tie is still a long way from cheap, and none of these improves the main bet you are already making.

Do scorecards and streak patterns help?

No, and the reason is more interesting than a flat no. The shoe does carry a trace of memory, because cards that have been dealt are not coming back, so baccarat is countable in the strict sense. It is just worth almost nothing. The Wizard of Odds reproduces Peter Griffin's result that a player keeping a perfect count and betting $1,000 only on the hands where the count has turned positive earns about 70 cents an hour, and its own analysis has a level-one count moving the banker bet from 1.06% to roughly 1.01%. A positive-expectation tie bet comes up about once every 475 hands. So the bead plate and the big road are decoration, and the honest version of the answer is not that the shoe has no memory but that its memory is worth less than the drink.

How does this compare to what real online baccarat software publishes?

Closely, on one bet, and the studio doesn't say which. Pragmatic Play publishes 98.76% for Baccarat (reconfirmed 2026-09-29, unchanged since first read), which lands on this page's computed player-bet return, 98.76%, almost exactly. It does not state which wager its single headline figure describes, which on a table with three main bets priced three different ways is exactly the ambiguity this page's bet-by-bet breakdown exists to remove. A second studio, Play'n GO, published a comparable figure for the banker bet until it withdrew the page that figure was read from; actfap.com/rtp no longer carries a Play'n GO comparison for that reason. actfap.com/rtp carries the Pragmatic Play figure, dated and linked to the provider's own page.

Methodology

How this is calculated

Every figure here comes from two published eight-deck probabilities: the banker hand wins 0.458597 of coups and the player hand 0.446247, which leaves 0.095156 for the tie. Nothing is simulated.

The banker bet pays even money less commission and pushes on a tie, so its expected value per unit staked is 0.458597 times one minus the commission, minus 0.446247, which is 1.06% against the house at the standard 5%. The player bet has no commission, so it is simply 0.446247 minus 0.458597, or 1.24%.

The tie pays 8 to 1 on a 9.5156% shot, which is 14.36%, and 4.84% where a casino pays 9 to 1.

The commission slider solves the same expression for whatever rate you set: the break-even point where the banker bet turns positive is one minus 0.446247 divided by 0.458597, or 2.69%, and the crossover where the player bet becomes the better of the two is 5.39%.

The commission-free variants are derived the same way from the probability of the event each one charges you on: a banker win with a three-card 7 on 2.2534% of coups for EZ Baccarat, and a banker win totalling 6 on 5.3864% of coups for Super 6.

The side bets are priced from their full published distributions rather than from a single event, so the 2.65% and 9.37% on the two Dragon Bonus wagers are summed out of the nine-row pay table each one settles on. The Panda 8 is 3.4543% at 25 to 1.

The pair bet is not transcribed at all: after the first card is dealt, 31 of the 415 cards left in an eight-deck shoe share its rank, and 31 divided by 415 at 11 to 1 is 10.36%.

Common misreadings5

The banker bet being the best bet on the table is a ranking, not a recommendation. It is the cheapest of the ways to lose money at this game, at roughly $18.51 an hour on a $25 flat bet, and where that sits against anything else you might do with the money is not a question arithmetic answers.
A commission-free table is not a table with the house edge removed. The rule that replaced the commission is where the money went, and on the Super 6 layout it went up rather than down.
The player Dragon Bonus being the cheapest side bet on the layout is not a reason to make it. It is a 2.65% bet sitting beside a 1.06% one, so adding it to a banker bet raises what the hand costs rather than lowering it. The reason it is worth knowing is the opposite one: that a bet's name tells you nothing about its price, and the identically named wager on the other hand costs about three and a half times as much.
EZ Baccarat's banker bet being marginally cheaper is not a reason to sit at an EZ table rather than a commission table. The four cents per $100 it hands back is smaller than one Dragon 7 bet gives away, and the Dragon 7 is why the layout exists.
A long run of banker or player results is not a pattern that has started, and it is not one that is due to end. It is the shape random data usually has, which the drill on this page is there to let you see for yourself rather than take on trust.

Assumptions5

Eight decks, shuffled and dealt under the standard drawing rules, with the tie pushing the banker and player bets. Six-deck and single-deck games shift the probabilities in the third decimal place, which moves the tie bet more than it moves banker or player.
The commission is charged only on winning banker bets, which is the near-universal rule. A house that charged it on every banker bet placed would be running a materially more expensive game than this slider models.
The Dragon 7 figure assumes the standard 40 to 1 payout. Side-bet paytables are set property by property, and a table paying less than that on the same event is a worse bet than the one priced here.
The other side-bet figures assume their standard schedules too: the Dragon Bonus paying 30, 10, 6, 4, 2 and 1 on winning margins of 9 down to 4 with a natural win at even money and a natural tie pushing, the Panda 8 at 25 to 1, and the pair bet at 11 to 1. Variants exist. A pair bet sold with a three-of-a-kind bonus on top is a different distribution and a different price, and is not what the 10.36% here describes.
The hourly costs quoted in the text use 70 coups an hour, the mini-baccarat pace this site uses in its house-edge comparison. A full big-baccarat table with three dealers and a card-passing ritual is considerably slower, and dealing speed changes the hourly cost in direct proportion.

Limits5

Every house edge here is an average over the long run. It is the right number for what a game costs to play and the wrong number for what any particular session will do, and baccarat's session-to-session spread is wide enough that the difference between banker and player will be invisible inside it for a very long time.
The edges are quoted per unit staked, the convention the published figures use. Because ties return the stake, a coup that actually settles costs more than the headline: 1.17% on banker and 1.36% on player. Both numbers are correct and they answer different questions. The side bets in the ranking are quoted on the same per-unit-staked basis, which is the right one for comparing them and is not adjusted for the Dragon Bonus pushing on a natural tie.
The probabilities treat coups as independent, which they are not exactly, because the shoe depletes. The error that approximation introduces is smaller than the last digit shown, which is the same fact that makes counting the game unprofitable.
The streak drill deals from those fixed probabilities rather than from a modeled shoe, so it is exactly memoryless. A real shoe is very slightly not, in a direction worth about 70 cents an hour to a perfect counter.
The drill's score is out of coups dealt, and a tie is a coup neither call takes, so it sits in that denominator and can never reach the numerator. The consequence is worth stating because it is not the one a reader expects: the long-run ceiling on the drill's headline figure is 45.86%, which is the banker win probability, and nothing beats it because calling banker on every coup is the best available call. Read over only the coups that settled, the same ceiling is 50.68%, which is the banker's share of decided hands this page opens on. The page prints both scales rather than one, so a reader finishing on 45% can see which number they are looking at instead of concluding they underperformed a coin flip.

Evidence

The two input probabilities are transcribed in math.ts from the Wizard of Odds baccarat basics page, retrieved 2026-09-07, with the source URL and date in the file header; that page publishes all three outcome probabilities to six decimals and the house edges they imply.

The commission-free probabilities come from the same site's EZ Baccarat page and its commission-free baccarat appendix, both retrieved 2026-09-09 and both cited in the same header. math.test.ts pins each of the four transcribed probabilities to the digit its source prints, then derives the 1.02%, the 7.61% and the 1.46% and checks each against the house edge that source publishes separately, so a mistyped digit fails rather than propagating quietly into a percentage nobody would notice was wrong.

It also asserts the claim the commission-free section exists to make, that the three banker bets rank EZ, then 5% commission, then Super 6, and that the ranking survives being recomputed per coup settled rather than per coup dealt.

The side-bet section is sourced the same way: the two Dragon Bonus distributions are transcribed row for row from the Wizard of Odds baccarat side-bets appendix, retrieved 2026-09-10, and the Panda 8 probability from the same EZ Baccarat page as the Dragon 7.

The 2.65%, the 9.37% and the 10.19% are then summed out of those rows and checked against the house edges that source publishes separately, and each Dragon Bonus table is checked to sum to a probability of exactly 1, which catches a mistyped digit that happened to leave the house edge looking plausible.

The pair bet's 10.36% is derived from 31 divided by 415 rather than transcribed, and the claim about why the two Dragon Bonus wagers differ is asserted against the tables themselves rather than only stated in the copy.

The card-counting figures in the FAQ are the Wizard of Odds analysis and its reproduction of Peter Griffin's result, and are cited rather than derived here.

The two ceilings the streak drill prints are derived from the same two input probabilities rather than transcribed, and math.test.ts pins both: that the best a caller averages is the banker win probability itself, that calling player or guessing averages less, and that the settled-coup version reproduces the 50.68% the intro states.

A passing test means the arithmetic matches the published figures, not that any casino near you deals these exact rules.

The whole-layout ranking (the intro's cost-per-unit-staked sentence, the FAQ's side-bet ranking and the misuse note comparing the two Dragon Bonus wagers) is read from layoutRanking(0.05) in math.ts, the same function the rendered table on this page builds from at whatever commission the reader's slider is set to, and every relational claim about that ranking (a factor of five and a half, three and a half times, about seven times) is computed from it by describeRatio rather than typed by hand a second time; math.test.ts pins the ranking's order and each of those ratios.

Gambling can cause serious harm. The National Problem Gambling Helpline is free, confidential and open 24/7, call or text 1-800-MY-RESET (1-800-697-3738), or chat at ncpgambling.org/chat.

Read up

Go deeper

GuideHouse Edge, RTP and What Casino Games Actually Cost YouWhat house edge means, how it differs from RTP and hold, why turnover matters more than your buy-in, and which casino games are cheapest and priciest to play.

Related tools