Prediction Market Arbitrage 2026: Do the Gaps Pay?
Prediction market arbitrage looks like the one free lunch in this business: the same event trades at 31 cents on Polymarket and 33 cents on Kalshi, so you buy both sides and keep the difference. We priced 15 matched pairs against both venues' live order books at 21:14 UTC on 12 November 2026, and 11 of them showed a gap worth having before costs. Then we read each market's own resolution text, and every edge above a single cent belonged to a pair that is not asking the same question. This guide prints all 15 pairs, the fee a gap has to clear, the size the book will actually take, and the five minute check that separates a real gap from a mirage.

Key takeaways
- Of 15 hand matched Polymarket and Kalshi pairs, 11 showed a positive gross edge on executable quotes. After reading both sides' resolution text, only two were the same question, and both offered 1.0 cent per contract.
- The largest gap was 26.8 percentage points, on whether Alice Weidel becomes German Chancellor. It is not arbitrage: Kalshi's contract requires government formation after the next federal election and runs to 26 August 2029, while the Polymarket contract covers a different window entirely.
- Fees decide this. Polymarket charges takers C x rate x p x (1 - p) at a category rate, and Kalshi charges roundup(0.07 x C x P x (1 - P)). Near 50 cents the two legs cost about 3.0 cents per contract, so a gap under 3 cents at mid prices loses money before you start.
- The one trade that looked real: 100 contracts on the GTA VI Blinding Lights market cost 99.00 dollars for a certain 100 dollar payout, so 1.00 dollar gross, minus 1.07 dollars of Polymarket fee and 1.53 dollars of Kalshi fee. Net, a loss of 1.60 dollars.
- Size caps even a real edge. Polymarket's book there held roughly 782 dollars of liquidity on 850 dollars of daily volume, so the trade tops out in the hundreds of dollars.
On this page
- What prediction market arbitrage actually means
- How we measured it, and why executable quotes matter
- What the 15 pairs showed
- The distribution, in three numbers
- Why the biggest gaps are not arbitrage
- The fee wall: what a gap has to clear
- The book runs out before the edge does
- When a cross venue gap is genuinely tradeable
- How to check a pair yourself in five minutes
- Frequently asked questions
- Sources and when we last checked
This is a plain list rather than jump links, because guide pages on this site render section text as paragraphs. If you read only two sections, read the one on resolution text and the one on fees. Between them they are the reason most advertised prediction market arbitrage does not survive contact with a real order book.
What prediction market arbitrage actually means
- Cross venue arbitrage: buy Yes on the cheaper venue, No on the dearer one, hold both to resolution. The one people mean.
- Intra venue arbitrage: buy Yes and No on one venue when the asks sum to under a dollar. Rare, because a single matching engine prices both legs.
- Direction one cost: Polymarket Yes ask plus (1 minus Kalshi Yes bid).
- Direction two cost: Kalshi Yes ask plus (1 minus Polymarket Yes bid).
- Your edge is one dollar minus the cheaper of those two, before fees. We report the cheaper direction for every pair.
Every contract on Polymarket and Kalshi pays one dollar if it resolves Yes and nothing if it resolves No. That single fact makes arbitrage thinkable: hold Yes on one venue and No on the other for the same event, and exactly one of them pays a dollar whatever happens. If the pair cost under a dollar, the difference is yours regardless of the outcome.
So the trade is not a bet on the event. It is a bet that two books disagree by more than the cost of trading them.
The arithmetic is worth writing down before you look at any screen. To buy Yes on venue A you pay A's ask. To buy No on venue B you pay one dollar minus B's Yes bid, because the No ask mirrors the Yes bid. Add the two: below a dollar is a gross edge, above it you are paying for the privilege.
How we measured it, and why executable quotes matter
- Source: Polymarket Gamma events and markets endpoints, Kalshi public markets and events endpoints. No keys, no authentication.
- Prices: bestBid and bestAsk on Polymarket, yes_bid and yes_ask on Kalshi. Executable quotes, not last prints.
- Timestamp: one pull at 21:14 UTC on 12 November 2026.
- Sample: 168 Kalshi markets with daily volume against 45,028 open Polymarket markets, hand verified down to 15 pairs.
Most writing on this subject quotes last traded prices, which is the main reason published arbitrage numbers look better than anything you can fill. A last print says what somebody paid at some point, not what is on the book now: a market whose last trade was 40 cents can have a best ask of 46 and nothing in between.
So we used top of book on both sides. Polymarket's Gamma API publishes bestBid and bestAsk; Kalshi's public markets endpoint publishes yes_bid and yes_ask. Those are prices an order would actually meet. Everything below comes from one pull at 21:14 UTC on 12 November 2026, so the pairs are comparable with each other rather than smeared across an evening.
Finding the pairs was the slow part. Kalshi's open markets list is dominated by multi leg parlay contracts: 12,000 consecutive open markets came back as combination tickers with no daily volume. The events endpoint instead gave 5,000 events, of which 168 single markets had traded in the previous day. On the Polymarket side we pulled 45,028 open order book markets.
We matched the two sets on question text, then checked every candidate by hand, which is how the false pairs got caught: the matcher proposed a Kalshi market on the next Israeli defence minister against a Polymarket market on the next Israeli prime minister. Fifteen survived. The sample is small on purpose, because the error being measured is the kind a matcher cannot see.
What the 15 pairs showed
- GTA VI radio, Blinding Lights. Kalshi 32.0/33.0, Polymarket 29.0/31.0. Gap +2.5pp. Cost 99.0c, gross edge +1.0c.
- GTA VI radio, God's Plan. Kalshi 22.0/25.0, Polymarket 22.0/24.0. Gap +0.5pp. Cost 102.0c, gross edge -2.0c.
- GTA VI radio, Mo Bamba. Kalshi 12.0/16.0, Polymarket 9.0/14.0. Gap +2.5pp. Cost 102.0c, gross edge -2.0c.
- GTA VI radio, Look at Me. Kalshi 12.0/15.0, Polymarket 10.0/11.0. Gap +3.0pp. Cost 99.0c, gross edge +1.0c.
- Next Bond, Robert Pattinson. Kalshi 31.0/32.0, Polymarket 14.0/16.3. Gap +16.3pp. Cost 85.3c, gross edge +14.7c.
- Next Bond, Callum Turner. Kalshi 35.0/36.0, Polymarket 31.5/33.2. Gap +3.1pp. Cost 98.2c, gross edge +1.8c.
- Next Bond, Jack Lowden. Kalshi 30.0/31.0, Polymarket 31.6/31.7. Gap -1.2pp. Cost 99.4c, gross edge +0.6c.
- Next Bond, Jacob Elordi. Kalshi 4.0/5.0, Polymarket 0.1/0.2. Gap +4.3pp. Cost 96.2c, gross edge +3.8c.
- Next Bond, Josh O'Connor. Kalshi 4.0/5.0, Polymarket 4.5/4.6. Gap -0.1pp. Cost 100.5c, gross edge -0.5c.
- Next Bond, Nicholas Hoult. Kalshi 8.0/9.0, Polymarket 2.8/4.2. Gap +5.0pp. Cost 96.2c, gross edge +3.8c.
- Next US House Speaker, Hakeem Jeffries. Kalshi 94.5/94.6, Polymarket 97.4/97.5. Gap -2.9pp. Cost 97.2c, gross edge +2.8c.
- Next Israeli Prime Minister, Gadi Eisenkot. Kalshi 68.0/74.0, Polymarket 64.0/64.4. Gap +6.8pp. Cost 96.4c, gross edge +3.6c.
- Next German Chancellor, Alice Weidel. Kalshi 28.0/35.0, Polymarket 1.7/7.7. Gap +26.8pp. Cost 79.7c, gross edge +20.3c.
- National Popular Vote compact reaches 270. Kalshi 41.0/47.0, Polymarket 52.0/53.0. Gap -8.5pp. Cost 95.0c, gross edge +5.0c.
- Next Millennium Prize problem, Riemann Hypothesis. Kalshi 0.0/4.0, Polymarket 2.6/6.2. Gap -2.4pp. Cost 101.4c, gross edge -1.4c.
Here is the whole sample, with nothing dropped for being inconvenient. Prices are in cents. The gap is the Kalshi mid price minus the Polymarket mid price, so positive means Kalshi was dearer. The cost is the cheaper of the two directions, and the edge is one dollar minus that cost, before any fee.
The distribution, in three numbers
Across the 15 pairs the median absolute mid price gap was 3.0 percentage points, the mean 5.7, the maximum 26.8. Eleven produced a positive gross edge in at least one direction, which on its face is a startling hit rate, and roughly what the promotional guides on this query promise.
The median matters more than the maximum. A measured study of 186 pairs on 30 August 2026 found a median fresh gap of 0.78 cents and a maximum of 7.5. Ours is wider because the sample skews towards thinly traded long dated questions, and thin books carry wide spreads. That is a property of the sample rather than a discovery.
Four pairs had a negative edge, and those are the honest ones: God's Plan, Mo Bamba, Josh O'Connor and Riemann all cost more than a dollar for both sides, which is what two competently priced books should produce. The question is what the other eleven are telling us.
Why the biggest gaps are not arbitrage
- Do both contracts share a deadline? Compare the resolution text, not the end date field, which is often stale.
- Does either side carry an extra qualifier, such as by AI, after an election, or at launch?
- What happens if the event never occurs? One venue may resolve No while the other has a none of the above outcome.
- Who resolves it, and from which source? Different settlement sources can split a genuinely identical question.
- Is a subset pricing above its superset? If so, stop: that is a thin book, not an edge.
We read the resolution text on both sides of every pair, which as far as we can tell nobody writing about this does systematically. The incumbent study flags pairs whose resolution dates sit more than three weeks apart, then tells the reader to check the wording. The wording is where the money is.
Every edge above one cent in our sample came from a pair that is not asking the same question. Not one was a pricing error. Three examples carry the argument.
The Robert Pattinson pair is the cleanest. Polymarket resolves Yes only if an actor is officially announced as the next James Bond by 31 December 2026 at 11:59pm ET, and resolves to No Bond chosen if nobody is announced in that window. Kalshi resolves Yes if Pattinson is cast before 1 January 2030. Kalshi is pricing three extra years of casting, so the apparent 14.7 cent arbitrage is two different questions, correctly priced. Putting the trade on buys you three years of time risk for nothing.
The National Popular Vote pair shows the tell you can spot without reading anything. Kalshi asks whether the compact reaches 270 electoral votes before 1 January 2030, at 41 to 47 cents. Polymarket asks whether it is implemented before the 2028 election, at 52 to 53. The shorter window trades above the longer one, which cannot be right: anything happening before 2028 has also happened before 2030. When a subset prices above its superset, one book is thin rather than wrong.
The Riemann pair is the narrowest trap. Kalshi asks which Millennium Prize problem will be solved next; Polymarket asks whether Riemann will be the next one solved by AI, a strict subset. The prices should differ, and any matcher keying on titles pairs them anyway.
The fee wall: what a gap has to clear
- At 5 or 95 cents: about 0.6 cents per contract across both legs.
- At 10 or 90 cents: about 1.1 cents.
- At 20 or 80 cents: about 1.9 cents.
- At 30 or 70 cents: about 2.5 cents.
- At 40 or 60 cents: about 2.9 cents.
- At 50 cents: about 3.0 cents, the worst case.
- All six assume Polymarket's 0.05 Culture rate plus Kalshi's 0.07, both as takers. A Polymarket Geopolitics leg is free, which roughly halves the wall.
Two pairs survived the resolution check as the same question: Blinding Lights and Look at Me, both asking whether a song is playable on an in game radio station in Grand Theft Auto VI at launch. Both offered a gross edge of 1.0 cent per contract. Now the fees.
Polymarket charges takers only, at shares times a category rate times p times (1 minus p), where p is the share price. Makers are never charged and earn a rebate. The rate is 0.05 for Culture, Sports, Economics and Weather, 0.04 for Politics, Finance and Tech, 0.07 for Crypto, and zero for Geopolitics. Both GTA VI markets sit in Culture, so 0.05 applies. Kalshi's published schedule charges roundup(0.07 x C x P x (1 - P)) on trades that take liquidity.
Run the trade. Buy 100 Yes on Polymarket at 31 cents for 31.00 dollars, and 100 No on Kalshi at 68 cents, the mirror of its 32 cent Yes bid, for 68.00 dollars. Outlay 99.00 dollars against a certain 100 dollar payout, so 1.00 dollar gross. The Polymarket fee is 100 x 0.05 x 0.31 x 0.69, or 1.07 dollars; the Kalshi fee is 0.07 x 100 x 0.68 x 0.32 rounded up, or 1.53 dollars. Net: a loss of 1.60 dollars on a trade that cannot lose.
Because both formulas run on p times (1 minus p), the cost peaks at 50 cents and falls towards both extremes. That shape tells you how big a gap must be before the resolution text is even worth reading.
The book runs out before the edge does
- Blinding Lights: Polymarket book about 782 dollars, daily volume 850 dollars; Kalshi open interest 5,547 contracts.
- Look at Me: Polymarket book about 885 dollars, daily volume 245 dollars; Kalshi open interest 1,906 contracts.
- Robert Pattinson: Polymarket book about 4,794 dollars; Kalshi open interest 287,694 contracts.
- Polymarket minimum order size is 5 shares, and the tick is 0.001 on the Bond markets and 0.01 on the GTA VI ones, so a one cent edge on a penny ticking market has no room to be improved.
- Capital is locked until resolution on both legs, which on a 2030 Kalshi contract is years rather than weeks.
Suppose a gap did clear the fee wall. The next question is how much of it you can have, and that is the number no competing page on this query states. A percentage return means nothing without the size it applies to.
On Blinding Lights, Polymarket's book carried roughly 782 dollars of liquidity on 850 dollars of daily volume, and Kalshi showed 5,547 contracts of open interest on 448 traded that day. Look at Me was thinner still, about 885 dollars of Polymarket liquidity on 245 dollars of daily volume. These are not markets where you deploy capital; they are markets where a 200 dollar order moves the price against you.
The long dated Bond markets are the opposite shape and no more use. Kalshi showed 287,694 contracts of open interest on Pattinson against about 4,794 dollars of Polymarket liquidity, so the deep side is the one you do not want and the shallow side caps the trade. Depth on one venue does nothing when both legs must fill.
When a cross venue gap is genuinely tradeable
- Both contracts resolve on identical wording, deadline and settlement source, verified by reading rather than by title.
- The gross gap beats the fee wall at that price, with room for the rounding.
- At least one leg can rest as a maker, or the gap is wide enough to pay both taker fees.
- Both books show live size at the quoted price.
- The resolution date is near enough that locked capital is not the real cost.
None of this means cross venue pricing is always efficient. The edge lives somewhere narrower than the guides suggest, and the conditions are specific enough to write down.
The strongest version is to be a maker rather than a taker on at least one leg. Polymarket never charges makers and pays them a rebate, and Kalshi's help centre says maker fees apply only on some markets, so two resting limit orders can turn a one cent gross edge into a real one. What you give up is certainty: a resting order may not fill, and a hedge with one leg on is a naked position, which is the risk the fee was buying off.
The second condition is a short horizon. A locked cent on a contract resolving next month is a different proposition from the same cent locked until 2030. Divide your edge by the years you will wait before calling it a return.
The third is that the gap must come from a fresh quote on both sides. You cannot trade against a price nobody is offering, and a wide spread on a market doing 245 dollars a day is not an opportunity but an absence of counterparties.
How to check a pair yourself in five minutes
- Open the Polymarket market, copy its slug, then fetch gamma-api.polymarket.com/markets?slug=THE-SLUG and read the description field in full.
- Fetch api.elections.kalshi.com/trade-api/v2/markets/TICKER for the Kalshi side and read rules_primary and close_time.
- Put the two resolution texts side by side. A different deadline, qualifier or settlement source means stop here.
- Only then compare bestBid and bestAsk against yes_bid and yes_ask, and compute both directions.
- Check the fee wall for that price band, then the book depth on the thinner side. An edge that survives all four is genuinely rare.
Everything above used two public endpoints and no credentials, so you can repeat it on any pair. The discipline that matters is the order of operations: read the resolution text first and the prices second. Doing it the other way round is how a 26 point gap becomes a plan instead of a warning, because once you have seen the number you will find reasons for it.
Frequently asked questions
Is prediction market arbitrage profitable?
Rarely, and far less often than the gaps suggest. In our 15 pair sample at 21:14 UTC on 12 November 2026, 11 pairs showed a positive gross edge, but every edge above one cent came from contracts that resolve on different terms. The two pairs that asked the same question offered 1.0 cent against a combined taker fee of about 2.6 cents. The honest version needs a resting maker order, a short horizon and a live quote on both sides.
Why do Polymarket and Kalshi prices differ on the same event?
Usually because it is not the same event. Deadlines, qualifiers and settlement sources differ far more often than titles suggest, and a difference in wording shows up as a difference in price. After that the common causes are thin books with wide spreads and stale quotes on one side. Genuine disagreement between two deep, fresh books on identical wording is the rarest explanation.
How do I calculate a cross venue arbitrage?
Add the price you would pay for Yes on one venue to the price for No on the other, which is one dollar minus that venue's Yes bid. Under a dollar, the difference is your gross edge per contract. Check both directions, since only one will be cheaper. Then subtract both fees: Polymarket charges takers shares times a category rate times p times one minus p, and Kalshi charges 0.07 times contracts times price times one minus price, rounded up.
What fees do Polymarket and Kalshi charge?
Polymarket charges takers only, at a category rate: 0.05 for Culture, Sports, Economics and Weather, 0.04 for Politics, Finance and Tech, 0.07 for Crypto, zero for Geopolitics. Makers are never charged and earn a rebate. Kalshi's published schedule charges roundup(0.07 x C x P x (1 - P)) on liquidity taking trades, with maker fees on some series only. Both peak near 50 cents, where two legs cost about 3 cents.
How big does a price gap need to be to arbitrage?
At mid prices, more than about 3 cents per contract if you take liquidity on both legs. Nearer the extremes the wall falls away, to roughly 1.1 cents at 10 or 90 cents and 0.6 cents at 5 or 95. Those figures assume Polymarket's 0.05 Culture rate and Kalshi's 0.07. If one leg is a Polymarket Geopolitics market the fee is zero on that side, which roughly halves the requirement.
Is arbitrage between Polymarket and Kalshi legal?
There is nothing unusual about holding offsetting positions on two CFTC regulated venues, and plenty of traders keep accounts on both. The practical constraints are access and funding rather than legality: your location decides which venue you can trade. Treat any profit as taxable. This is general information, not legal or tax advice.
Why do arbitrage scanners show bigger gaps than this?
Most read last traded prices rather than the order book, and a last print is not a price you can fill. A measured study of 186 pairs on 30 August 2026 found that none of its twelve largest raw gaps survived a check that both quotes were fresh within a day. Scanners also match on titles, the error that produced our Alice Weidel and Riemann pairs. Treat a scanner hit as a question, not an answer.
Sources and when we last checked
- Polymarket Gamma API, events and markets endpoints, pulled 12 November 2026: bestBid, bestAsk, liquidity, daily volume, the 5 share order minimum, tick sizes of 0.001 and 0.01, category tags and the resolution descriptions quoted above. The events endpoint now requires gzip for any request with a limit above 50.
- Kalshi public API, markets and events endpoints, pulled 12 November 2026: yes_bid, yes_ask, open interest, daily volume, close times and the rules_primary text for every Kalshi market named here.
- Polymarket fee documentation at docs.polymarket.com/trading/fees, read 12 November 2026: the taker formula, the per category rates, the zero rate on Geopolitics, and that makers are never charged.
- A measured cross venue study of 186 pairs priced on 30 August 2026, used for the 0.78 cent median fresh gap, the 7.5 cent maximum, and the finding that none of its twelve largest raw gaps survived a freshness check.
Two transparency notes, because they affect how much weight to put on the fee numbers. Kalshi's fee schedule PDF returned HTTP 429 to every automated fetch while this was written, so the roundup(0.07 x C x P x (1 - P)) formula comes from indexed copies of that document, labelled effective 5 February 2026, corroborated across several independent summaries. Kalshi's help centre, which we did read on 12 November 2026, confirms the fee basis, that maker fees apply only on some markets and that cancelling a resting order is free, but does not print the constant.
One source conflict is worth naming rather than hiding. Third party guides disagree on Kalshi's maker constant, commonly given as 0.0175, and on whether rounding applies per order or per fill. Those details move a two leg calculation by fractions of a cent, which matters when the whole edge is one cent, so we have left the maker constant out of the figures above.
Our sample is 15 pairs at one timestamp, not a longitudinal study, and it skews towards thinly traded long dated questions. It shows that the large gaps are resolution mismatches rather than edges, a claim about kind rather than frequency, and cannot say how often a genuine gap appears on the busiest markets. Nothing here is financial advice. One line to carry away: on this trade the resolution text is the product, and the price is only a summary of it.