Kalshi–Polymarket Arbitrage: 32,638 Price Checks, and Where a Gap Survived Fees

A price gap between Kalshi and Polymarket is easy to find and hard to be paid for. Vultax matched the same outcomes on both venues, NFL, college football and Premier League games, the midterms and the September Fed decision, and walked both order books every five minutes for 28 hours on 12 and 13 September 2026: 32,638 checks of whether buying Yes on one venue and No on the other would pay after both venues' fees, at the size the books actually held.

The slower markets were where gaps paid. Midterm contracts cleared fees at depth in 10.1% of readings, and two pairs did so in every reading, the best paying $347.69 on money that stays locked until the election is called. The steadiest edge of all was a quirk of tick sizes on a Fed longshot, worth 0.18% of the capital it needed. Before kickoff, sports almost never paid: pre-game NFL quotes did not cross once in 7,518 readings, college football cleared fees once, for 21 cents, and the Premier League never did.

Every one of those edges is small against the capital, the wait and the risk that the two venues' rules resolve the same event differently. The calculator runs the arithmetic on any two prices.

Matched contracts must describe the same outcome under the same settlement conditions. Start with the full rules and identifiers on both venues. Similar titles are only a way to find candidates. Polymarket's resolution documentation distinguishes the market question from the rules that determine its payout.

For a Yes-on-one-venue and No-on-the-other comparison, write down what each leg would pay in every relevant case. A pair is an exact match only when the combined payout remains the assumed $1 per pair. Keep near matches out of that calculation. The Fed example below shows why a rate threshold and a particular size of cut are different contracts.

Use the contract-review worksheet to keep the two rulebooks and edge cases together. Then use the gap calculator for the arithmetic. Neither a completed worksheet nor a positive calculation proves that both orders can fill.

• Outcome and side: compare the same candidate, team, threshold or rate bucket, and confirm that Yes and No refer to complementary outcomes.
• Clock: compare the event window, cutoff, time zone and date used for the result. An event time and a settlement deadline can differ.
• Evidence: record each venue's resolution source and the treatment of revisions or disputed results.
• Exceptions: test ties, postponements, cancellations and replacement candidates against both rulebooks.
• Prices and size: after matching the rules, compare executable asks for the same quantity, with an observation time for each book. A midpoint is not a fill.
• Costs: enter the applicable fee on each leg, rounding and transfer costs separately. Record any unfilled quantity and how long capital remains committed.

Every five minutes from 18:51 UTC on 12 September to 22:51 UTC on 13 September, Vultax read Kalshi's public trade API and Polymarket's public market and order-book endpoints: 337 cycles. Outcomes were matched across venues by teams and date for games, by race and party or candidate for the midterms, and by rate bucket for the Fed, then checked against both rule texts.

For each matched pair and each cycle, Vultax took the cheaper venue's Yes asks and the other venue's Yes bids, which are the same thing as No asks, and walked them level by level, charging each venue's taker fee on each level: the rate Polymarket publishes for that market, 0.05 for sports and economics and 0.04 for politics, and 0.07 for Kalshi, the rate independent trackers report for its standard fee. It stopped when the next level no longer cleared both fees. Sports pairs count only readings taken at least ten minutes before the scheduled start.

Crossed means one venue's ask was below the other's bid before fees. Cleared at depth means at least one contract could be bought on both sides at a profit after fees.

Six midterm pairs and one Fed pair account for every reading that cleared fees at depth outside a single college-football moment. Neither kind is forced shut quickly: the midterm contracts settle months away, so closing a gap ties up capital until then, and the Fed gap sits on a tick boundary that Kalshi cannot quote through.

• Senate control - a 0.91% return is roughly 6% a year if the media call the Senate in early November, and about 2.4% a year if settlement waits for Kalshi's 1 February 2027 fallback, against the 4% or so that cash earns.
• The Fed tick gap - Kalshi quotes whole cents and Polymarket tenths of a cent, so a longshot bid at 1 cent on Kalshi and offered at a fraction of a cent on Polymarket crosses by construction. The depth sat at Kalshi's 1-cent bid, and the return is 0.18% over four days, before Kalshi rounds its fee up to the cent.
• Maine - the two venues write the contract differently: Kalshi pays if a Republican is sworn in, Polymarket if Susan Collins wins. A hedge across them is only a hedge if both happen together; the midterms comparison sets out the rules.

Before kickoff, NFL books were deep on both venues: the median Kalshi game market held more than $1.6 million within two cents of its midpoint between one and 24 hours before the start, and Polymarket's held $280,000 to $500,000. Measured over the six hours before kickoff instead, the medians were $1.74 million on Kalshi and $523,000 on Polymarket, as Vultax's study of sports and esports books shows. With one-cent spreads and that much size on each side, a gap large enough to clear two taker fees did not survive five minutes once. College football books were thinner on Polymarket, and that is where nearly all pre-game crossings appeared, at a median of one cent before fees and below zero after them.

During games the picture looks different and is not trustworthy. In 6,296 in-play readings, 16.9% showed crossed quotes and 7.9% appeared to clear fees. But each cycle read every Kalshi book before every Polymarket book, so the two sides of a pair could be up to two minutes apart while the score changed. Those readings are left out of every figure above. Measuring live-game gaps needs both books captured within the same second.

A pair that pays $1 in every outcome on paper can pay nothing if the rules part company. For the games in this study the core result matches, but the edge cases differ.

A gap only means something if both prices were read at the same moment. In 295 paired reads of seven events carried on both venues, taken hourly on 17 and 18 September 2026, the two venues' order books arrived a median 1.1 seconds apart and never more than 1.85. A two-second matching rule keeps every read.

A one-second rule keeps 100 of the 295, and what it drops is whole events: every read of the Israeli prime minister, Brazilian first-round and Nobel Peace Prize markets, and 47 of 48 reads of the December Fed meeting. The gap follows the number of books in the event: six Senate-control books came back 0.07 seconds apart, 38 Nobel books took 1.76. A study that matches on a tight clock ends up measuring only the small events.

Polymarket's October contract asks how far the upper bound of the target range moves. Kalshi's asks where it ends up: KXFED-26OCT-T3.75 resolves Yes when the upper bound finishes above 3.75%. Its No side covers every outcome at or below 3.75%, which is more than one size of cut.

Take a starting upper bound of 4.00%, chosen for illustration. Cuts of 25 and 50 points both pay Kalshi's No. Only one of them pays Polymarket's 25-point cut. One outcome in four pays differently, which is enough to reject pairing the two prices one to one. Kalshi lists a second October threshold at 4.00%, so the same meeting name covers several different level bets.

The last constraint is legal, not arithmetic. The Polymarket prices in this study come from its international exchange, which U.S. residents cannot use, while Kalshi is a U.S.-regulated exchange; where U.S. law stands explains the split. A gap that only one group of traders can reach from each side can persist for reasons that have nothing to do with anyone missing it. Vultax did not measure how much of the midterm gaps that explains.

Vultax's cross-venue record measured the same question earlier, on eight Fed-rate contracts matched across the two venues and observed roughly every 85 seconds for 24 hours from 17 July 2026: 8,144 observations. It recorded quoted prices and the liquidity available at them rather than walking each book level by level, and it counted a gap only if it cleared a conservative threshold that charged taker fees on both legs, with slippage and operational buffers.

On the one pair whose terms matched exactly, Polymarket's 4.50% end-2026 upper bound against Kalshi's KXFED-26DEC-T4.25, the median gap was 574 basis points, about 5.7 cents, but the average size available at the quoted prices was $14. Across all eight pairs, 509 of the 8,144 observations, 6.3%, cleared the threshold, every one of them on that exact pair, where they were half of its 1,018 observations. Three quarters, 6,108 observations, were blocked before fees because the two contracts were near matches rather than the same bet, and 17 of the 19 gaps Vultax tracked were still open when the window ended.

Measured in different months and in different ways, the two readings agree: the rule text and the size in the smaller book decide more than the width of the gap. On that exact pair the 90th-percentile gap was 934 basis points, and five of the six near-match pairs sat between 128 and 352. How the two venues differ in structure, fees and what they hold is set out in what Kalshi and Polymarket hold.

Order books were read to 20 levels on Kalshi and 20 on Polymarket, every five minutes; a gap that opened and closed between readings is not counted, and one that persisted is counted in every reading, so the shares describe how often a gap was visible, not how many separate chances there were. Matching was checked against both venues' rules; outcomes that could not be matched unambiguously were left out, including many college-football games whose names differ between venues. Polymarket fees are the rates published for each market; Kalshi's 0.07 coefficient is the rate reported by independent trackers because its fee schedule could not be retrieved from our environment, and its rounding is ignored in the walk. Capital is the cost of both legs plus fees. Deposits, withdrawals, the cost of holding capital, and a second leg that fails to fill are not included.

Nothing here is a recommendation to trade. Challenge a figure at [email protected]; material changes are logged in the changelog.

The downloadable Kalshi-related input and derived data files are temporarily unavailable while source redistribution permissions are reviewed. Methods and limitations remain described here; contact [email protected] for questions. This is not a claim that Vultax owns the underlying venue data.

The July reading is one 24-hour window, eight Fed-rate pairs and one exact match, read from Vultax's registry of Polymarket markets paired with Kalshi tickers, each pair graded exact, near or excluded by a reviewer. It shows how matching decides what a gap means; it is not a market-wide statistic.