Polymarket vs. Kalshi: Fees, Settlement and Access

By Alex Thornton · author of the MIT-licensed teachersbettextbook library · Last reviewed 24 July 2026

Same object, two machines. Both trade a claim that pays $1 or nothing, and both run an order book — then they diverge on the three things that actually cost you money: the fee formula, who decides the outcome, and who is allowed in the door.

Polymarket and Kalshi both run central limit order books on binary event contracts, but they differ in three ways that matter: how fees are charged, who determines settlement, and who may legally trade. Kalshi is a CFTC-regulated exchange settling in dollars; Polymarket settles in USDC against an on-chain oracle.

The fee formulas, on one contract

Price the identical position at both venues: 100 contracts at 60¢, crossing the spread. Kalshi’s published taker schedule:^[1]

Polymarket’s published schedule has a different shape: a base rate applied to the cheaper side of the contract.^[2]

where:

• $C$ — contracts (Kalshi) or shares (Polymarket). Same object: a $1-or-nothing claim.
• $P$ — the trade price in dollars.
• $r$ — Polymarket’s base fee rate, a venue-set parameter. It has long been reported as zero on the central limit order book, which is why “Polymarket has no trading fees” is the usual shorthand.

So the honest side-by-side, on this contract, as of the review date: $1.68 against $0.00. That is a real difference and it is also the least stable number on this page — a base rate is a dial the venue can turn, and any figure here is a dated snapshot, not a guarantee. Check the venue’s own fee page before you rely on either.

Why the shapes matter more than the levels

Levels change. Shapes are structural. Kalshi’s fee is a quadratic hump in the price; a rate on $\min(P, 1-P)$ is a linear tent. Both peak at a coin flip; they part company everywhere else, and the ratio has an exact closed form, because $P(1-P) = \min(P,1-P) \times \max(P,1-P)$:

Set a hypothetical $r = 2\%$ purely to make the shapes comparable — the ratio is then $3.5 \times \max(P, 1-P)$, and it grows toward the tails:

Both fees are cheap in absolute cents at the tails; the quadratic is relatively dearer there. And on either venue the number that decides anything is fee relative to outlay: on Kalshi that is $0.07(1-P)$, which is 6.7% of the money at a 5¢ contract and 0.7% at 90¢. Cheap in cents, expensive in percent.

Three fine points that fee tables tend to omit. Kalshi charges a maker roughly a quarter of the taker rate, with a flat 0.25% maker fee on some major events — trivial at 50¢, a real slice of a sub-5¢ contract.^[1] Kalshi rounds the fee up to the next cent, which on a 1-contract order at 90¢ turns 0.63¢ into a full penny, 59% above rate. And on Polymarket the costs that survive a zero fee are the spread and the on- and off-ramp — moving dollars into and out of USDC is not free, and it is a per-trip cost that does not appear in any fee formula.

Depth, ticks and how orders rest

Structurally the two books are close cousins: resting bids and asks, price-time priority, makers and takers, and the identity that buying NO at 40¢ is selling YES at 60¢. Where they differ:

The mint-and-redeem row is the mechanically interesting one: because a dollar of collateral can always be split into a YES and a NO and recombined, $P_{\text{YES}} + P_{\text{NO}}$ is pinned near $1 by construction rather than by convention. Both venues are read the same way once you have the four numbers — bid, ask, and the depth behind each — and on both, the midpoint is the lazy answer. The microprice is the right one.

Settlement: the difference that has no analogue at a sportsbook

Kalshi. Each contract ships with a written rulebook naming the settlement source and the resolution criteria, and the exchange determines the outcome against it, under CFTC oversight, with a defined process for disputes. Settlement itself is free — a round trip pays the fee twice, holding to resolution pays it once.

Polymarket. Resolution runs through the UMA optimistic oracle. A proposer posts the outcome with a bond; a challenge window opens; if nobody disputes, that answer stands. If somebody does dispute, the question escalates to a vote of UMA token holders.^[3]

Read that last clause slowly, because it is a risk category that does not exist on a regulated exchange or at a sportsbook: ambiguous market wording is resolved by token-holder vote. It is fast and cheap when a question is crisp. When a question is loosely worded, the outcome depends on how a governance process reads the sentence, and there is a public history of contested resolutions on exactly that boundary. For a fair-value model this is not a footnote — a contract whose settlement is contestable is not the same instrument as one whose settlement is not, even when the two trade at identical prices.

Access, as of 24 July 2026

A dated snapshot, because this is the fastest-moving row in the table and any page that states it flatly is wrong within a quarter.

• Kalshi operates as a CFTC-regulated designated contract market. Several state gaming regulators have taken the position that sports event contracts are wagering under their statutes; exchanges have contested those orders in federal court on preemption grounds, rulings have gone different ways in different courts, and the question was live litigation at the time of this review.
• Polymarket has historically restricted US persons following a US regulatory settlement, and its US posture has been changing. Whether, and on what terms, US users may trade is a question for the venue’s current terms — not for a betting-math site.
• Betfair Exchange is not available to US users at all, which is why it appears in this cluster as a consensus member under review rather than as a venue we quote.

This is where the competing content in this niche fails hardest: it states all of the above as resolved, usually with an affiliate link attached. It is not resolved. Nothing on this page is legal or tax advice, and event-contract tax treatment in particular is not obviously the same as gambling-winnings treatment — ask a professional about your own situation.

Two venues, one opinion

Here is the part that matters if you are building a fair line rather than picking a venue. Kalshi and Polymarket list the same events, and traders move capital between them, so any meaningful gap gets closed:

Because that pressure is constantly applied, the two venues’ prices — and their errors — move together. Treating them as two independent opinions and averaging is a statistics error with a measurable size. For two estimators of equal variance $\sigma^2$ and correlation $\rho$:

So the pooling step discounts their combined weight by the measured residual correlation, and the grading step is leave-one-out: when scoring a venue’s price against the benchmark, that venue is excluded from the benchmark. Skip that and you measure a venue partly against itself and systematically understate the error. It is not a refinement; it is a correctness bug.

Consistent with the rest of the site, the method above is published and the fitted numbers are not: the measured correlation between venues and the learned per-venue weights stay private, while the board publishes the result — fair probability, confidence, agreement, venue count. The same boundary applies to the sportsbook pool.

Sources

1. Kalshi. “Fee Schedule” (July 2026 revision). — taker and maker formulas, the round-up-to-the-cent rule, the flat 0.25% maker case, and the zero settlement fee.
2. Polymarket. Developer documentation — central limit order book, fee schedule and market mechanics. — the base-rate-on-min(P, 1−P) fee shape; the rate is venue-set and should be read live.
3. UMA. Optimistic Oracle documentation. — propose, bond, challenge window, escalation to a token-holder vote.
4. Burgi, C., Deng, W. & Whelan, K. (2026). “Makers and Takers: The Economics of the Kalshi Prediction Market.” GWU Working Paper 2026-001. PDF. — informative prices, favorite–longshot bias after fees: neither venue is an unbiased oracle.
5. Genest, C. & Zidek, J. V. (1986). “Combining Probability Distributions: A Critique and an Annotated Bibliography.” Statistical Science 1(1), 114–135. — pooling estimates, and why correlated sources cannot be counted twice.

Every formula here lives on the Formula Sheet for quick reference.