Prediction Market Arbitrage: Kalshi vs Polymarket vs PredictIt

Find prediction market arbitrage between Kalshi, Polymarket and PredictIt. This actor matches the same Yes/No question across the three venues and prices the hedge: buy YES on one venue and NO on the other. Together those pay exactly $1 whichever way the question resolves, so when they cost less than $1 after fees, the difference is locked-in profit.

Each matched pair comes with:

• The arbitrage: which venue to buy YES on and which to buy NO on, at their live asks, the total cost, fees, and net profit and return. Rows with real arbitrage sort to the top.
• The odds gap: both venues' YES probability, the spread, and whether that gap clears both order books.

No account, no API key, no browser. Straight from each venue's public API.

Arbitrage fields

Field	Meaning
arbitrage	true = YES + NO cost under $1 after fees, both venues quote both sides of their book, and the pair is not a suspected mismatch
arbStrategy	Plain language, e.g. "Buy YES on kalshi at 0.42 + buy NO on polymarket at 0.55 = 0.97 for a $1 payout"
arbYesVenue / arbYesAsk	Where to buy YES and the ask you pay
arbNoVenue / arbNoAsk	Where to buy NO and the ask you pay
arbCostPerContract	YES ask + NO ask
grossProfitPerContract	$1 minus the cost
feesPerContract	Kalshi taker fee (0.07 x P x (1-P)) and PredictIt's 10% profit fee, counted on both legs to be safe. Polymarket charges none on most markets.
netProfitPerContract / netReturnPct	Profit after fees, per contract and as a return on the cost
suspectedMismatch	The pair is probably two different questions, so it is not marked as arbitrage. That covers: YES prices more than 25 points apart, a number other than a year on one side only (a district like NJ-04, a date like Jan 1), or a match score under 0.8.
bothBooksTwoSided	Both venues quote YES and NO. A one-sided book is usually a stale order you can't fill.

Set onlyArbitrage: true to get only the opportunities, and minNetReturnPct to skip the small ones. Schedule it every few minutes to catch gaps when they open.

The gap that is not a gap

A mid-price difference between two venues is only worth acting on if it survives crossing both order books. Thin contracts routinely quote bid/ask tens of points wide, and a mid-to-mid "gap" inside that width is not tradeable at any size.

Every row therefore carries each venue's own quoted width and a verdict:

Field	Meaning
widestBookWidthPct	The wider of the two venues' YES bid/ask spreads, in points
gapExceedsBookWidth	true = the gap clears both books, false = it sits inside the spread, null = a venue did not quote both sides

In a representative run, 17 of 25 matched pairs had gaps sitting inside the quoted bid/ask. Filter on gapExceedsBookWidth: true to keep only the pairs where the edge is real.

What you get

One row per matched cross-venue pair:

Field	Meaning
question	The matched question
venueA / venueB	Which two venues this row compares
venueAYesPct / venueBYesPct	Implied YES probability on each side of this pair (0-100)
spreadPct	Absolute gap between the two YES prices, the headline number
cheaperYesVenue	Which venue prices YES lower
gapExceedsBookWidth	Whether the gap clears both venues' bid/ask (see above)
buySignal	Plain-language read, including when the gap is not tradeable
matchScore	Match confidence (0-1) between the two questions
kalshiYesPct / polymarketYesPct / predictitYesPct	Per-venue YES price. The venue not in this pair reads null.
kalshiBookWidthPct / polymarketBookWidthPct / predictitBookWidthPct	Each venue's own quoted bid/ask width
kalshiVolume / polymarketVolume24h / predictitVolume	Liquidity. predictitVolume is always null: PredictIt publishes no volume field, and null means not reported rather than zero traded.
predictitContract	Which contract inside the PredictIt market this leg is
predictitBestBuyYesCost / predictitBestSellYesCost	The raw YES ask and bid, null when that side of the book is empty
kalshiCloseTime / polymarketEndDate / predictitDateEnd	Resolution timing. PredictIt leaves this unset on most contracts, so it reads null.
kalshiUrl / polymarketUrl / predictitUrl	Links to the underlying markets

Column blocks for all three venues are present on every row, so the shape stays stable no matter which pair produced it.

Input

Field	Description
venues	Which venues to compare. Pick at least two; a shorter selection falls back to all three.
searchQueries	Topics to compare (e.g. president, minister, trump, fed). Empty = most-traded markets on every selected venue.
minSpreadPct	Only return pairs whose YES prices differ by at least this many points.
minMatchScore	How strict the cross-venue match must be (0-1, default 0.5). Raise for precision, lower for recall.
minKalshiVolume	Skip thin Kalshi markets (lifetime contracts).
minPolymarketVolume24h	Skip thin Polymarket markets (24h USD).
poolPerVenue	Markets scanned per venue before matching (default 4,000).
onlyArbitrage	Only rows where arbitrage is true.
minNetReturnPct	With onlyArbitrage, the minimum net return in percent.
maxPairs	Cap on pairs returned.

Volume filters do not apply to PredictIt, which reports no volume.

How matching works

Each question is reduced to a token set (stopwords dropped, common synonyms and month names normalised, $150,000 becomes 150000). A pair is scored by the overlap coefficient (shared tokens divided by the smaller set) and only kept when it shares at least one strong token (a number or a word of 4+ letters), so pairs never form on filler words alone.

The venues structure the same race differently, and the matcher is built for that. Kalshi lists one binary market per candidate ("Will Marine Le Pen win the 2027 French presidential election?"); PredictIt nests a contract per candidate inside one market ("Who will win the next French presidential election?" plus "Marine Le Pen"). Both are expanded into individual Yes/No legs before matching, so the candidate name carries the match even when the wording differs.

Matches are assigned greedily, best score first, one market per side per venue combination, so the Kalshi leg of a question can still pair with both Polymarket and PredictIt. Use matchScore to filter and eyeball the per-venue titles before trading.

Pricing

Pay per event: $0.005 per matched pair pushed. The first 2 rows of every run are free, and a run that finds no matches emits a single free note row. Nothing else is charged.

Notes & limits

• PredictIt is US politics only and is a small market with per-contract position limits, so a gap against it may not be fillable at size even when gapExceedsBookWidth is true.
• PredictIt contracts are independent Yes/No bets. The Yes prices inside one PredictIt market do not sum to 1, because the contracts are not mutually exclusive. Each is treated as its own binary question, which is what makes it comparable to Kalshi and Polymarket at all.
• Works best on named-entity event markets (elections, geopolitics, courts, awards) where the venues list the identical yes/no question. Crypto price levels are structured differently on each venue (Kalshi uses price ranges, Polymarket uses thresholds) and rarely match cleanly.
• Prices are live snapshots at scrape time; all three venues move continuously.
• A high spreadPct can also mean the two questions differ in a detail (date, threshold) the matcher missed. Confirm the titles and resolution terms before acting.
• Check before you trade. Arbitrage is only locked in if both markets resolve on the same event. Read both venues' resolution rules: dates, sources and edge cases can differ even when the titles match.
• Prices are top-of-book asks. Depth isn't shown, so a large order can move past them. PredictIt also caps each contract at $850 and charges 5% on withdrawals, which isn't counted here.