Gaps and arbitrage
Real Kalshi-Polymarket gaps right now in 2026
Most days, none. On 2026-09-21 TapeXray scored 2 fee-adjusted gaps between Kalshi and Polymarket out of 6 matched pairs, with 100 more pairs rejected before scoring because the venues were not asking the same question. The live count on this page is re-scored every deploy, and it is usually lower.
What this page found
- TapeXray counts 2 real fee-adjusted Kalshi-Polymarket gaps on the board timestamped 2026-09-21 17:30 UTC.
- The same gate has 4 matched pairs inside the fee line, so the visible spread is not enough after fees.
- The match process rejected 100 suspect pairs before scoring because the contracts were not the same question.
- Positive pairs are 33.3% of the gated set, with 2 positive after fees and 4 inside the fee line.
- The archive tracks 236,027 markets across two venues as of 2026-09-21.
How many real Kalshi-Polymarket gaps are there right now?
The current answer is 2 real fee-adjusted gaps. That is the number on the live board after TapeXray matches a Kalshi contract to a Polymarket contract, checks that the question is close enough to score, and subtracts the board fee estimate. The board is timestamped 2026-09-21 17:30 UTC.
The number is small because most apparent differences are not trades. Some are just inside the fee line. Some compare contracts that sound similar but do not resolve on the same event. Some are wide on the midpoint but not fillable at the quoted edge. Most days there is nothing here. On this board there is something, but it is still only 2 gaps across a tracked catalog of 236,027 markets.
What gets rejected before a gap becomes real?
The board has 6 gated pairs in the scored set: 2 positive after fees and 4 inside the fee line. That is why the positive share is 33.3%, not because a third of the full archive is arbed. The full archive is much larger, and the match process is deliberately harsh.
The bigger number is 100 suspect pairs rejected before scoring. That is where many public arb counts go wrong. A mayoral election market can differ on office, date, body, or resolution source. A sports market can differ on overtime, abandoned matches, league stage, or settlement rule. A politics market can differ on who certifies the result. The tape does not treat those as free money. It throws them out.
- Inside fee line: 4 pairs where the spread does not beat the fee estimate.
- Positive after fees: 2 pairs where the spread survives the board fee estimate.
- Suspect rejected: 100 pairs removed before scoring because the match was not clean.
- Gated positive share: 33.3%, calculated from the 6 gated pairs on the board.
Where are the matched pairs on the board?
The matched-pair category table is not the arb count. It is the labeled match inventory used by the board. The rows below total 106 matched pairs: 5 + 13 + 1 + 33 + 29 + 21 + 3 + 1 = 106. Only a smaller gated set reaches the fee-adjusted scoring step for the current board.
Politics and Sports supply much of the clean matching work, but Other is the largest bucket on this timestamp. That is normal for a broad prediction-market archive. Contract wording varies more outside standardized sports and major-election markets, so the category mix does not translate directly into arb supply. Readers who want the raw market surface can start from the markets archive and then use the gap board for scored cross-venue pairs.
| Category | Matched pairs |
|---|---|
| Crypto | 5 |
| Culture | 13 |
| Economy | 1 |
| Other | 33 |
| Politics | 29 |
| Sports | 21 |
| Tech | 3 |
| Weather | 1 |
What was the last real gap?
The last real gap on the board is the Berlin mayor market: "Will Elif Eralp win the election for Governing Mayor of Berlin by the Berlin House of Representatives following their state election?" The board shows a 3.0c fee-adjusted gap, with the Kalshi midpoint at 81.0c and the Polymarket midpoint at 73.5c.
The second listed gap is the 46th FIDE Chess Olympiad Open Tournament market on Uzbekistan, with a 2.1c fee-adjusted gap, a Kalshi midpoint of 60.0c, and a Polymarket midpoint of 51.5c. These are board readings, not promises of execution. A trader still has to check asks, bids, limits, settlement text, and whether both legs can be filled before the market moves. The related fee treatment is explained in Kalshi vs Polymarket fees.
| Market | Category | Kalshi mid | Polymarket mid | Fee-adjusted gap |
|---|---|---|---|---|
| Elif Eralp Berlin mayor | Politics | 81.0c | 73.5c | 3.0c |
| Uzbekistan 46th FIDE Chess Olympiad Open | Other | 60.0c | 51.5c | 2.1c |
How do fees change the arb count?
Fees are the difference between a pretty midpoint chart and a bettable line. Kalshi charges takers with the formula 0.07 x P x (1 - P) per contract per side, before its cent-rounding rule in the published schedule. At 50c, the raw fee shape is 0.07 x 0.5 x 0.5 = 0.0175 dollars, or 1.75c per contract per side.
Polymarket publishes category rates rather than one flat rate. Crypto is 0.07, sports is 0.05, politics is 0.04, and geopolitics is 0.0. The live gap board does not use that full category schedule in its midpoint gap line. It charges the Polymarket leg a flat 6% taker estimate for gap math. That 6% is a TapeXray approximation, not Polymarket’s published fee schedule. For why scanner counts can be too high, see why arb scanner numbers are wrong.
| Venue | Taker fee basis | Maker fee | Verification |
|---|---|---|---|
| Kalshi | 0.07 x P x (1 - P) per contract per side | 0.0 | 2026-09-22 |
| Polymarket | Category rate x P x (1 - P) | 0.0 | 2026-09-22 |
| TapeXray board gap line | 6% Polymarket taker estimate inside board math | Not a venue fee | 2026-09-21 17:30 UTC |
Questions people ask
Does 2 real gaps mean there are 2 guaranteed arbitrages?
No. The board count means 2 Kalshi-Polymarket pairs are positive after the board fee estimate at 2026-09-21 17:30 UTC. Execution still depends on bids, asks, limits, timing, settlement wording, and both legs filling. The count is a filter for work, not a guarantee of profit.
Why were 100 suspect pairs rejected?
TapeXray rejected 100 pairs before scoring because the two contracts were not clean matches. Similar wording is not enough. If one venue’s market has different settlement rules, timing, event scope, or resolution source, the pair can show a fake spread that is not a real arbitrage.
Why are 4 pairs inside the fee line?
The board shows 4 gated pairs with a visible spread that does not survive the fee estimate. In bettor terms, the price difference is not enough to beat the vig. Those pairs may still be useful for line shopping, but they are not counted as real fee-adjusted gaps.
What fee does the board use for Polymarket?
The live board uses a flat 6% Polymarket taker estimate in its gap math. That is a TapeXray approximation. Polymarket’s published schedule uses category rates, including 0.07 for crypto, 0.05 for sports, 0.04 for politics, and 0.0 for geopolitics.
How big is the archive behind this count?
The catalog tracks 236,027 markets across two venues as of 2026-09-21. That includes 97,194 Kalshi markets and 138,833 Polymarket markets. The real-gap count is much smaller because only cleanly matched, fee-adjusted pairs reach the scored board.
Sources
Independent archive. No referral links in editorial copy, no paid placements, and nothing here is legal, tax or financial advice. Found an error? Tell us and we will date the correction.
Related reading
Get told when a gap opens
Most days the board is empty, which is exactly why watching it by hand is a waste of your evening. The alert channel pings when a pair clears the fee line.
Free Telegram alerts. Example:
Get alerts (opens in a new tab)