MATH OF LUCK

Low pair or high cards: the video poker hold, priced on the 8/5 table

Illustration for “Low pair or high cards: the video poker hold, priced on the 8/5 table”
On this engine’s 8/5 Jacks or Better table, the expected value of each hold, computed by enumerating every possible draw: a low pair (sixes) returns 0.81 units against 0.49 for keeping a king and a queen instead; a pair of jacks returns 1.53 against 1.06 for breaking it to draw at a four-flush; a made flush returns 5.00 against 1.31 for three to the royal; and four to the royal returns 19.45, which beats keeping a made straight at 4.00 but not a made straight flush at 50. Two pair (2.51) beats holding either pair alone (0.82); a full house (8.00) beats holding the trips (4.25). None of these are opinions: the deck has 47 unseen cards and the draws are countable, so each number is the average payout over every draw, weighted equally. The gap between the right hold and the tempting one is typically 0.3 to 0.5 units per hand, which on a game that returns about 97.3% at best is the difference between the published price and a much worse one.
BETKYO RESEARCHPUBLISHED 2026-09-13UPDATED 2026-09-138 MIN READ

How a hold is priced

After the deal you can see five cards and 47 are unseen. Whatever you keep, the replacements are drawn from those 47 without replacement, and every combination of them is equally likely. So the value of a hold is not a matter of feel: keep two cards and there are 16,215 ways the other three can come; keep four and there are 47. Add up the payout of every resulting hand, divide by the number of ways, and you have the expected return of that hold in units of the stake. Do it for each way of holding and the best one is the correct play.

ENGINE-VERIFIEDThe figures in this article were computed for it by exact enumeration: for each hold, every combination of replacement cards from the 47 unseen was dealt, the final hand was ranked, and it was paid from VP_PAYTABLE in vpoker/derive.ts — royal flush ×800, straight flush ×50, four of a kind ×25, full house ×8, flush ×5, straight ×4, three of a kind ×3, two pair ×2, jacks or better ×1. The engine header states the table’s return as about 97.3% at optimal play. No simulation is involved; the averages are exact.

One consequence is worth having in mind before the table. The engine draws replacements from a sequence that was fixed when the round was committed, in the order described in the cursor article, so the cards that arrive are not a response to your choice. But the price of each choice is the average over all the sequences that could have been committed, and that is what the numbers below are.

The classic dilemmas, priced

Expected return of each hold, in units of the stake, on the engine’s 8/5 table
DEALHOLDRETURNSTHE TEMPTATIONRETURNS
6♠ 6♦ K♥ Q♣ 9♠The pair of sixes0.81King and queen0.49
J♥ J♦ 4♥ 8♥ K♥The pair of jacks1.53Four hearts to the flush1.06
A♥ K♥ 2♥ 7♥ Q♠Four hearts to the flush1.09Ace, king, queen0.46
10♥ J♥ Q♥ 5♥ 9♥The made flush5.00Three to the royal1.31
10♥ J♥ Q♥ K♥ 9♠Four to the royal19.45The made straight4.00
10♥ J♥ Q♥ K♥ 9♥The made straight flush50.00Four to the royal18.47
5♠ 5♦ 9♥ 9♣ K♥Both pairs2.51One pair, drawing three0.82
K♠ K♦ K♥ 8♣ 8♦The full house8.00The three kings4.25
9♠ 10♦ J♥ Q♣ 2♥The open-ended straight draw0.81Jack and queen0.49
2♠ 7♦ 9♥ J♣ K♥Jack and king0.49Draw five new cards0.32

Every row is the exact average over all draws. “Returns” includes the stake: 1.00 means you get your bet back on average.

Three of these are the ones players get wrong most often, and the table says why each is wrong by a specific amount. The low pair looks like nothing and the two honours look like something, but the pair is worth 0.81 and the honours 0.49: the pair reaches trips, two pair, full houses and quads often enough to outweigh the honours’ frequent small save at ×1. A pair of jacks is a paid hand, and breaking it for a four-flush trades 1.53 for 1.06, a loss of almost half a unit; the flush comes in 9 draws out of 47, and ×5 is not enough to pay for the 38 misses. And three to the royal against a made flush is the romantic one: the flush pays 5.00, the chase returns 1.31, and the romance costs 3.69 units every time.

The two royal rows show where the chase does pay. With four to the royal, one of the 47 cards completes an ×800 hand and several others still make a flush or a straight, so the draw is worth 19.45 units, nearly five times a made straight. It is still not worth a made straight flush: ×50 in hand beats 18.47 on the draw, and the correct play is to keep the 9♥ and take the money.

What the 8/5 table changes

Every number above was computed on the table this engine deals, which since 10 September 2026 is 8/5: full house ×8, flush ×5. The full-pay 9/6 table it replaced paid one unit more on each of those rows, and the paytable article explains what that costs overall, about two points of return. What it does to the holds is smaller than people expect. The flush draw in row three is worth 1.09 here and would be worth about 1.17 on a 9/6 table; the pair of jacks in row two is worth 1.53 here and slightly more there, because a pair sometimes becomes a full house. The ranking of the holds does not change on any row in the table. The price of the game changed; the correct play did not.

That is the general shape of video poker strategy: the paytable sets the ceiling, the holds decide how close you play to it, and the two are mostly independent. A player who learned the ranking on 9/6 loses nothing by keeping it on 8/5. A player who reads the 8/5 label and decides the strategy no longer matters loses roughly a third of a unit every time a low pair sits next to two honours.

Two things the table does not say. The house edge is not beaten by any hold: the best play on this table returns about 97.3% of turnover, and every row above is a way of losing less, not of winning. And the engine cannot see your holds before it deals: the replacement cards were derived from the committed seed pair before you chose, which is why each of these averages is the honest price of the decision and not a story about what the machine wanted.
FAQ

Should I keep a low pair or two high cards in video poker?

Keep the low pair. On this engine’s 8/5 Jacks or Better table a pair of sixes next to a king and a queen returns 0.81 units on average; keeping the king and queen returns 0.49. The pair reaches trips, two pair and full houses often enough to beat the honours’ occasional ×1 save.

Should I break a pair of jacks for a four-card flush?

No. The pair returns 1.53 units; drawing at the four-flush returns 1.06. The flush arrives in 9 of 47 draws and ×5 does not cover the 38 misses.

Should I break a made flush for three to the royal?

No. The flush is worth 5.00 units in hand; three to the royal returns 1.31 on the draw. Four to the royal is different: it returns 19.45, which beats a made straight (4.00) but not a made straight flush (50).

Does the 8/5 paytable change the correct holds?

Not in any of the classic cases. It lowers the value of flush and full-house outcomes slightly, so a few marginal decisions elsewhere shift, but the ranking of every hold in this article is the same as on the 9/6 table. The table changes the price of the game, not the strategy.

How were these numbers computed?

By exact enumeration for this article: every combination of replacement cards from the 47 unseen was dealt for each hold, the resulting hand was paid from the engine’s VP_PAYTABLE, and the payouts were averaged. There is no simulation and no rounding beyond the two decimals shown.

SOURCES & REFERENCES
  • Betkyo engine source: VP_PAYTABLE (8/5, ×800 royal, ~97.3% at optimal play), deriveVpokerSequence and vpFinalHand in vpoker/derive.ts
  • Exact hold enumeration written for this article against that paytable (10 deals, every hold, every draw from the 47 unseen cards)
  • Wizard of Odds — Jacks or Better strategy and paytable analyses
THE GAMES IN THIS ARTICLE
Betkyo Research — written by the team that builds these games. Every probability quoted in the Journal is derived from our engine source or a cited reference, never copied from another site. Figures are re-checked whenever the engines change.

18+ · PLAY RESPONSIBLY · THE JOURNAL IS EDITORIAL CONTENT, NOT BETTING ADVICE