Polymarket US charges 2.72% of cash and Kalshi 1.52%

From October 21 to 23, 2026, Polymarket US kept $2.7188 out of every $100 of cash —price × contracts: what the buyer pays— traded on its public tape, while Kalshi kept $1.5241 in a sample of 48,000 trades from Oct-23: 1.78 times more. And the fee schedules for both are nearly identical. Both charge a quadratic fee —contracts × price × (1 − price), multiplied by a constant their documentation calls Θ— and that constant has been 0.0695 at Polymarket US since Sep-17-2026 and 0.07 in the Kalshi fee schedule from July 2026: Kalshi's is 0.72% higher. What separates the two figures is where the money sits on the price curve: for every dollar of cash, the average Kalshi contract is bought at 78.22 cents and the Polymarket US contract at 60.95 — in a fee structure that vanishes as the price approaches one dollar, Polymarket US's mix is 1.79 times Kalshi's. Both figures come from primary sources: 8,835,082 trades from the Polymarket US tape and the Kalshi public API. The 0.07 constant comes from the Kalshi fee schedule text obtained by proxy and the DefiLlama open adapter; its API confirms the quadratic form and series multipliers, but not the constant. Furthermore, the Kalshi fee row in the aggregator everyone consults has been stalled for five months.

How much do Polymarket US and Kalshi really charge for every dollar traded?

A prediction market contract pays $1 if the event occurs and $0 if it does not, so there are two ways to measure volume, differing by a factor of five. Cash (cash or premium in both platforms' documentation) is price × contracts: what the buyer puts up. Notional counts contracts as if they were all worth one dollar. The average Polymarket US contract is bought at 19.24 cents, so its cash is 19.24% of its notional. Both measures are correct; the one that comes out of the user's pocket is the former.

The convention row is included because Kalshi's denominator has a catch. Its API returns every trade with the "yes" side price, and 36.75% of the sample are takers —those who cross against an order already on the book— buying the "no" side: 17,638 out of 48,000. Their documentation is explicit that buying "no" at 20 cents is the same position as selling "yes" at 80, and that whoever goes long on "no" pays the "no" price. Charging those trades at the "yes" price inflates cash by 22.5% and lowers the fee per dollar to 1.2443%; at the price of the side actually bought, it comes to 1.5241%. In dollar terms, the fee is identical in both readings: p × (1 − p) is symmetrical.

Why is there no single "fee" for a prediction market?

In a standard crypto exchange, the fee is a percentage of the amount. In Kalshi and Polymarket US, it depends on the contract price. The formula in the Kalshi fee schedule, literally, is round up(M × 0.07 × C × P × (1−P)), with C contracts, P price, and M a series multiplier. The Polymarket US formula, as registered by QCX LLC —the entity operating the market— with the CFTC, the US derivatives regulator, is Θ × C × p × (1−p) with Θ = 0.0695.

Divide that fee by the cash (C × p) and you get Θ × (1−p): the percentage of what you put up depends only on the price, and it grows as the contract gets cheaper. In Polymarket US, a contract at 5 cents pays 6.60% of the amount spent; at 50 cents, 3.48%; at 99 cents, 0.07%, fifty times less than at 50. From end to end, the curve sweeps across nearly two orders of magnitude under the same fee schedule. Per contract, the shape is different: there, the maximum is at 50 cents, at 1.74 cents per contract.

The two columns are almost identical, with Kalshi always a hair higher. An average platform fee cannot be read directly from the schedule: one must weight that curve by the cash flowing at each price. That weighting —the expected value of (1 − price), weighted by cash— is 0.3905 at Polymarket US and 0.2178 at Kalshi. Multiplied by the constant, it yields 2.7142% and 1.5246%, accounting for almost everything measured: at Polymarket US, 0.0046 percentage points are missing —0.17% of its fee— contributed by combos with their additional term.

Why does Polymarket US charge 1.78 times more than Kalshi with a nearly identical rate?

Because of composition, and the ratio factors into two parts. The Polymarket US constant is 0.71% lower than Kalshi's, which pushes in the opposite direction of the headline: a factor of 0.9929. The price mix is 1.7930 times higher. The product of the two is 1.7802 —compared to the 1.7839 times measured— with the difference made up by combos.

The two average prices in the table seem to contradict each other, but they do not. Per contract, both markets trade equally cheap items: 19.24 cents and 17.63. Per dollar of cash, 60.95 and 78.22. Many cheap contracts carry little weight in dollar terms, so the average price of money stays well above the average price of the contract: when Kalshi users put up cash, they put it into contracts closer to resolution, where the quadratic fee barely applies anymore.

The tape breaks it down by price deciles, using cash from the three days.

26.71% of Polymarket US cash passes through the $0.40 to $0.60 range, where the curve charges between 2.8% and 4.2% per dollar. The highest decile, $0.90 to $1.00, concentrates 19.76% and pays less than 0.70%.

How much does Kalshi collect daily compared to Polymarket US?

More than Polymarket US, despite charging less per dollar. Kalshi moved $1,543.1 million in cash over the three days according to the DefiLlama series, and Polymarket US $266.8 million according to its own tape: 5.78 times more (using the volume the aggregator attributes to Polymarket US, $255.0 million, it would be 6.05 times). A proprietary extrapolation —applying the Oct-23 sample ratio to the non-sampled Oct-21 and Oct-22, and using the 1.2443% aggregator convention because its volume series uses the "yes" side price— puts Kalshi's revenue in the range of $19.20 million over three days, compared to the $7.25 million the fee schedule yields on the Polymarket US tape: 2.65 times more in absolute terms.

The comparison reverses on the maker side, the one who leaves the order on the book. Polymarket US pays the maker a rebate —a fee refund— in tiers from -0.0035 to -0.0104 since Oct-19-2026 at 13:00 ET. Kalshi does not charge a maker fee by default, as its multiplier is zero, though it does in a short list of series.

Which Kalshi series do not pay the standard fee?

The Kalshi public API publishes the fee type and multiplier for each series, data that does not appear on any dashboard. A direct reading from Oct-25-2026, after the figures were finalized, returns 15,393 series: 15,198 with a quadratic fee and multiplier of 1, and 195 outside that mold, or 1.27%.

Those 195 fall into three blocks, with one counting twice because a baseball series is in two at once. Fourteen have a zero multiplier, meaning no fee: nine in politics and elections, three in economics, and two in crypto, KXBTCY and KXETHY, the Bitcoin and Ethereum annual close contracts. Nineteen have a 0.5 multiplier —half fee— and all nineteen are MLB baseball: RBIs, home runs, stolen bases, five-inning results. And 163 also charge a maker fee —0.0175 × C × P × (1−P), a quarter of the taker fee—: 107 in sports, 21 in science and technology, 13 financial, 10 in economics, 7 in entertainment, 3 exotic, and 2 in crypto, three of them with the parlay variant. In the Oct-23 sample, 157 of the 48,000 trades fell into non-standard series, and the maker fee was 0.10% of the total collected.

Why is the Kalshi fee series on DefiLlama useless since May-27-2026?

Because it stopped there. Its last data point is May-27-2026 and it sums to zero in the October window, while the volume series remains up to date (last point, Oct-24-2026). The summary endpoint does not return "no data": it returns total7d = $37,895,137, the exact sum of the last seven days with data, from May 21 to 27.

Anyone dividing that field by October volume gets a number that measures nothing. Using the seven days from Oct-14 to Oct-20 in the denominator yields 1.01%; with the three days from Oct-21 to Oct-23, 2.46%. Both are correct arithmetic over windows of different lengths. With the numerator frozen, the ratio simply measures how many days the person dividing has chosen. The last full week with data, May-18 to May-24-2026, gives 2.69% per dollar using the aggregator's own figures —double the October measurement— and as of today, it is impossible to tell if the difference is due to rates, mix, or series definition; the following week drops to 1.04% because it only has three days of fees against seven days of volume.

For Polymarket US, however, the aggregator is useful: its series is live and its methodology documents that it applies Θ × C × p × (1−p) to the tape. Over the same three days, it gives 2.6943% compared to the 2.7188% from the recalculation, with a 5.60% difference in fees and 4.65% in volume, and the gap is concentrated in one day: -2.2% on Oct-21, -1.7% on Oct-22, and +19.1% on Oct-23, when the tape shows $95.53 million in cash and the aggregator $80.21. Both sources cut by UTC day, so there is no time zone shift to explain it, leaving it open. In the Oct-14 to Oct-20 window, the difference was 0.02% in fees and -0.51% in volume. This type of error was previously discussed in how to read DefiLlama open interest: the aggregate field looks like a measurement but is just a sum of whatever is available.

Why did the Polymarket US fee per dollar rise 26% in seven weeks?

Because both components rose simultaneously, and the 40.6(d) self-certification filings that QCX LLC submits to the CFTC allow them to be separated. The aggregator ratio went from 2.1658% in week 35 (Aug-24 to Aug-30) to 2.7314% in week 42 (Oct-12 to Oct-18): a 26.11% increase. The two effects multiply rather than add: the constant contributes a factor of 1.1583 by rising from 0.06 to 0.0695, and the mix contributes another 1.0887 —the implicit weighting of (1 − price) went from 0.3610 to 0.3930— and 1.1583 × 1.0887 = 1.2611. In a longer perspective, the fee per dollar has gone from 1.82% in week 21 to 2.73% in week 43, which is incomplete as the aggregator only reaches Oct-24. Combos —trades tying multiple outcomes into one contract— are 0.48% of cash and 0.89% of the fee: they pay 1.85 times their weight.

What are the limits of the Kalshi and Polymarket US fee recalculation?

There are five, and they should be kept in mind before citing the figures.

1. Kalshi's is a sample, not a census. It consists of 48,000 trades totaling 0.56% of the day's cash under the aggregator convention ($2.90 million out of $514.6 million in its series). The design introduces a possible bias: each block of 2,000 trades covers between 5.6 and 36.9 seconds —median 12.3— just before the close of each UTC hour, totaling about five minutes of the day. The Oct-20 sample gave 1.5060%: the two readings with the corrected base differ by 1.2%, compared to the 10.2% that separated them under the old convention (1.3709% and 1.2443%).
2. Both figures are lower bounds. Rounding the recalculation to the nearest cent moves the result by -0.03% (1.2443% vs 1.2446% unrounded). What keeps the figure lower is the Kalshi model, which adds a per-order rounding fee to the schedule fee with a refund of the accumulated excess: "Net fee = trade fee + rounding fee − rebate (always ≥ $0.00)". And at Polymarket US, table tennis moved to Θ = 0.10 on Oct-7, but the tape symbols do not carry categories, so those trades are calculated at 0.0695.
3. The 0.07 Kalshi constant was not verified on their website nor can it be dated. I could not open kalshi.com directly, so the original fee schedule was not consulted there; the text I do have ("Fee Schedule for July 2026 — 7.7.26 Update") and the DefiLlama open adapter agree on 0.07 and 0.0175; the API, read without intermediaries, confirms the model shape (quadratic) and multipliers. The document says "last updated and effective: July 7, 2026," which does not confirm the constant changed that day.
4. The two windows do not match. The Polymarket US tape covers UTC days Oct-21 to Oct-23 and the Kalshi sample only Oct-23. For that single day, Polymarket US charged 2.8201% —$2,693,936.69 on $95,526,330.50 of cash— and the ratio with Kalshi rises to 1.85.
5. The side-bought correction can be applied to Kalshi but not to Polymarket US. Its API identifies the side of each trade; the Polymarket US tape does not, so the recalculation there uses the registered price for all, just like the aggregator's adapter. The two bases are not exactly the same convention, and the data freeze does not allow for an estimate of the value of that difference.

How will fees change after the Oct-28 FOMC and the Nov-3 elections?

The perishable part of these figures is the mix, and two dates will move it. On Oct-28-2026, the FOMC decides on rates and October contracts resolve: as a probability approaches 0 or 1, cash migrates to the high decile —the $0.90 to $1.00 range, which pays less than 0.70% per dollar— and the average fee drops without anyone touching the fee schedule. On Nov-3-2026, with the midterm elections, electoral volume kicks in, which at Kalshi falls partly on political series with a zero multiplier (nine of the fourteen without fees) and at Polymarket US spreads cash across the entire curve.

Sources and links: Polymarket US — daily time-and-sales files · docs.polymarket.us — fees (Oct-19-2026) · CFTC — QCX LLC, 40.6(d) from Sep-21-2026 (Θ 0.06 → 0.0695) · CFTC — QCX LLC, 40.6(d) from Jul-7-2026 · CFTC — QCX LLC, 40.6(d) from Sep-29-2026 (combos) · CFTC — QCX LLC, 40.6(d) from Oct-13-2026 (table tennis) · Kalshi — public trades API · docs.kalshi.com — order direction (bid/yes, ask/no) · docs.kalshi.com — fee rounding · Kalshi, July 2026 fee schedule (not linked: the server did not respond to our connections) · DefiLlama — Kalshi fee summary · DefiLlama — Kalshi fee adapter. Recalculations, weightings, and deciles: CleanSky calculation based on Polymarket US tape (UTC days Oct-21 to Oct-23-2026) and Kalshi API (samples from Oct-20 and Oct-23-2026), frozen on Oct-25-2026.