actfap

ICM Calculator

Turn tournament chip stacks into real money equity, and see how much a chip lead is actually worth.

This prices the chips that are on the table right now against the prize structure, so it values a stack for a deal or a single all-in decision, and it knows nothing about who plays better, where the button is, or how large the blinds are about to become.

ICM equity vs chip chop

$140.04

moves off chip share under this ladder, 12.5% of the $1,120.00 pool. Cut the structure to one paid place and it reads $0.00.

Prize pool
$1,120.00
Chips in play
105,000
Places priced
4

Players and stacks

  • ICM equity$354.61

    Gap−$125.39

    chop $480.00

    42.9% chips → 31.7% money

  • ICM equity$305.35

    Gap−$14.65

    chop $320.00

    28.6% chips → 27.3% money

  • ICM equity$249.24

    Gap+$57.24

    chop $192.00

    17.1% chips → 22.3% money

  • ICM equity$210.80

    Gap+$82.80

    chop $128.00

    11.4% chips → 18.8% money

Flattening the ladder moves more, not less

  1. Winner-take-all$0.00

  2. 900 / 120 / 60 / 40$46.68

  3. 700 / 250 / 120 / 50$79.16

  4. 500 / 300 / 200 / 120(shipped default)$140.04

  5. 280 / 280 / 280 / 280$240.00

Concentrated at the top, flat at the bottom, over this page’s own four stacks (45,000, 30,000, 18,000 and 12,000) into a $1,120.00 pool, the same at every row above. The transfer rises as the ladder flattens, never the other way.

The gap column is what ICM does: it shifts equity off the big stacks (who can only win once) and onto the short stacks (who ladder up when others bust). That gap is the reason to fold thin spots near a pay jump. One paid place is the only structure where the two divisions agree at every table, which is why the figure above prices the pay ladder rather than anyone’s stack. Neither is money anyone owes.

Written and reviewed by Alexander Korsager · last reviewed October 6, 2026 · corrections

The idea

What this tool shows

In a poker tournament, chips are not money. Once the prize pool is fixed, doubling your stack does not double your equity, because you can only win first place once and the pay jumps below it are worth a lot. The Independent Chip Model puts a dollar figure on a stack by assuming your chance of finishing first equals your share of the chips, then removing the winner and repeating for second, third and so on.

  1. The practical consequence is that chips you can lose are worth more than chips you can win.A big stack at the final table is worth less than its raw chip percentage; the short stacks are worth more than theirs. That is why correct tournament play tightens up near pay jumps and why the chip leader can apply so much pressure, everyone else has more to lose.

  2. This calculator runs the full model for final-table-sized fields.Use it to value a stack, to check a proposed deal against each player's real equity, or to understand why folding a marginally profitable spot in chips can still be the right call in dollars.

FAQ

Questions players ask

What is ICM used for?

Two things mainly: negotiating final-table deals so each player gets close to their true equity, and making end-game decisions, knowing when a call that is +chip-EV is −$-EV because of the risk to your tournament life near a pay jump.

Why is the big stack worth less than its chip percentage?

Because it can only finish first once, and the money saved by the field busting is spread among everyone. ICM redistributes some of the leader's notional equity to the shorter stacks, who are more likely to ladder up.

What are the limits of ICM?

It ignores skill, position, blind pressure and future play, it only sees stack sizes right now. It is also better at some of its own outputs than others. The chance of finishing first is exactly right in a game where chips only move between players, while the places below that are approximations, and in the one setting where the true answer can be computed they lean toward overpaying the short stack. Good enough to anchor a deal, not a complete model of tournament value. The methodology below has the numbers.

Methodology

How this is calculated

The model assumes one thing and derives everything else from it: your chance of finishing first is your share of the chips in play. Second place comes from removing a hypothetical winner, renormalizing the stacks that are left and asking the same question again, third from repeating that, and so on down the paid places.

A seat's equity is then the sum, over every possible finishing order, of that order's probability times the prize it pays, which is why the equities on the page always add up to exactly the prize pool you typed in. None of this is originally a poker result.

It is Harville's rule for multi-entry competitions, published for horse racing in the Journal of the American Statistical Association in 1973 (volume 68, pages 312 to 316), which is why poker's name for it, Malmuth-Harville, carries his.

The chip chop shown beside it is the naive comparison: each stack's raw share of the pool, with no pay ladder in it at all. The figure the panel leads with is neither of those two divisions but the distance between them.

It sums every seat that gains under ICM, and that sum is the whole transfer rather than half of it: both divisions pay out the same pool, so the dollars the short stacks gain are exactly the dollars the long stacks lose. It reads $0.00 in two cases.

One paid place is the first, because a single prize makes ICM equity chip share times the prize, which is the chip chop. A table where every stack is equal is the second, because both divisions are then an even split whatever the structure looks like.

Common misreadings4
  • A bigger figure at the top of the panel means a flatter pay ladder, not a steeper one, and that is the opposite of the obvious reading. Concentrating the pool into first place pulls ICM equity back toward chip share, so winner-take-all moves $0.00. Paying every seat the same makes finishing position worth nothing, hands everyone an even split, and is the furthest any division gets from chip share. The calculator on this page renders the full curve, from winner-take-all to a flat split, over the stacks and pool it ships with, so if a structure gets flattened during a deal you can read directly how far this figure moves rather than guessing. A flat ladder is ICM's widest disagreement with chip share, not its narrowest.
  • The gap column is not money anyone owes you. It is the distance between two ways of dividing the same pool, and the only division that pays out is the one every player at the table actually agrees to. At the shipped defaults the short stack reads $82.80 above the chop, which is this model's opinion about a stack, not a claim on anybody.
  • A seat worth more than the money it has already locked up is not a seat that is winning. At the defaults the short stack holds 11.43% of the chips and prices at $210.80 of a $1,120 pool, which is 18.8% of the money. That premium is the pay ladder rewarding survival, and it is paid for by finishing last 48.7% of the time.
  • Equity below your chip share is not a problem to solve. The leader at the defaults holds 42.86% of the chips and prices at $354.61 against a $480.00 chip chop, 26% under it. That is the model doing the thing it exists to do, and it happens at every final table with a pay ladder. It says nothing about how this particular table is going.
Assumptions5
  • First place is chip share, and that is the assumption rather than a finding. Every other number here is derived from it. It is exactly right in a game where chips only move between players and nobody has an edge, which is the reason it is the sturdiest part of the model and the places below it are not.
  • The first payout box is first place. The boxes are read in the order you type them and are not sorted, so a structure entered out of order is priced out of order, and any box left at zero is dropped rather than treated as an unpaid place. Dropped means the prizes under it move up a place: typing 500, 0, 200, 120 prices 200 as second and 120 as third, which is not the same table as one where second pays nothing.
  • Nobody can finish below the last seat, so a structure listing more paid places than there are seats prices only the places a seat can actually reach. The prize money in the places below that is left out of the pool, and out of every equity and chip-chop figure measured against it, not spread across the seats that remain. The panel names how many places and how much money that excludes whenever it happens.
  • Everybody is still playing for every place. No deal has already been struck, there are no bounties, no rebuys, and no chips arrive from outside the table.
  • Stack size is the only thing that separates two seats. Two players with the same chips have identical equity here, whoever they are and wherever they are sitting.
Limits4
  • Below first place the model is measurably wrong, and it is wrong in a consistent direction. Take the one case where the true answer can be computed exactly rather than estimated: three players, fair coin flips, one chip moving at a time, no skill. That is the setting Persi Diaconis and Stewart Ethier study in Gambler's Ruin and the ICM (arXiv:2011.07610, revised 2021), where their conclusion is that the model "is shown to be inadequate". Solving that chain exactly for stacks of 6, 4 and 2 units, both it and this tool put the leader's chance of winning at 50.00%, and then they part company. The short stack finishes second 21.86% of the time in truth and 25.00% of the time under this model. On a 500/300/200 structure that is $275.00 of equity against a true $271.86, and on a 500/300/0 structure, where the pay ladder is steeper, $158.33 against $148.92. In every stack and payout combination checked, the model overpaid the short stack and underpaid the leader.
  • That benchmark is not poker and it is not a correction you can apply to your final table. It has no blinds, no position and no skill in it, and blinds in particular cut the other way, since a short stack paying a fixed cost every orbit is not in a fair game in their favor. What it establishes is that the model has a known bias in a clean case, not the size or even the direction of its error at a real table.
  • Nothing about the future is an input. Skill, position, stack-to-blind ratio, a blind level about to jump, and who is sitting to your left all move real equity, and this page cannot see any of them. That is the gap between a number that anchors a negotiation and a number that settles one.
  • Ten seats is the ceiling, and the reason is arithmetic rather than caution. The calculation walks every finishing order, so the work grows factorially with the paid places: four seats paying four places is 64 branches, ten seats paying four places is 5,860, and ten seats paying all ten places is 9.86 million. Beyond that the honest answer is a different model, not a slower run of this one.
Evidence

The recursion is covered by math.test.ts, which pins it against hand-computed two- and three-player cases, checks that the equities always sum to the prize pool, and checks that more chips never price lower.

The transfer figure at the top of the panel is pinned in the same file: its two zero cases, that summing the seats which gain equals summing the seats which lose, and that a flat ladder moves more than a steep one, the last of those there so that no sentence on this page can drift into implying the reverse without failing the build.

The four structures compared above were measured against that module for this page rather than estimated. A passing test means the code computes Malmuth-Harville correctly. It does not mean Malmuth-Harville is correct, which is what the limits above are about.

The exact three-player figures quoted there were computed for this page from the Markov chain the Diaconis and Ethier paper describes, rather than transcribed from it.

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.

Related tools

Try next

We’d like to use privacy-friendly analytics (Google Analytics, Vercel Analytics and Speed Insights) to see which tools and pages are actually useful, so we can improve them. Nothing is sold, and nothing loads unless you say yes. See our privacy policy for details.