A mispricing between Kalshi and Polymarket lasts about 10 seconds

When Kalshi and Polymarket price the same game, their two lines rarely sit exactly on top of each other. A gap opens, one side moves, and the prices come back together. A question we get from people building on the feed is how long that gap stays open before it closes. We can answer it from our own record, because the matcher watches both books at once and timestamps each gap as it opens and again when it closes.

Over one 24-hour window, from 02:52 UTC on 7 July to 02:52 UTC on 8 July 2026, we logged 870 cross-venue gaps that closed on their own, meaning the two prices converged rather than the game ending and settling them. Half were gone within about nine seconds. Here is the shape of it.

The window covered a full slate. A World Cup round-of-16 match between Argentina and Egypt was in there, alongside Coco Gauff against Jessica Pegula in the WTA and Mexico against the United States in FIBA basketball. Most of the 870 gaps came from further down the card: a Tuesday night of MLB games, a long tail of ITF and Challenger tennis, and esports series in CS2 and Valorant that no sportsbook prices at all.

Measure	Value
Gaps observed	870
Window	02:52 UTC 7 Jul to 02:52 UTC 8 Jul 2026
Median time open	9.2 seconds
Closed within 10 seconds	55%
Closed within 30 seconds	96%
Longest single gap	106 seconds
Widest gaps, over 5% after fees	86 gaps, median 13.7 seconds
Median gap size, net of fees	$0.42

The curve above reads as the share of those 870 gaps still open at each second after the gap opened. It falls off a cliff. Nine seconds is the halfway mark, and by thirty seconds all but four percent have closed. The single longest gap in the window held for under two minutes, and it was an outlier. Size did not buy much more time either. The gaps wide enough to clear more than five percent after fees, 86 of them, ran to a median of 13.7 seconds before they closed.

Dollar size is a different measure than time, and it skews hard. Net of fees, the median gap in the full 870 was worth about 42 cents. The 86 widest gaps, the ones clearing five percent, ran to a median of $2.15. A handful of outliers pull any average up from there: the single largest gap in the window was worth $1,364.69. Most of what shows up in this dataset would not be worth executing even with a direct line to both books; the outliers are rare.

A window this short is not one a person can catch by hand. By the time you have opened both venues and lined up the same game across their two different tickers, the median gap has already closed. Seeing these moments at all means watching both books continuously and holding the two prices side by side as they move. That is the work the matched feed does, and it is why the gap between two public prices stays invisible unless you are watching both at the same instant.

A note on what is counted here. These are gaps the matcher confirmed as the same outcome on both venues, in its confirmed-parity band. We hold back a separate review band where the match is less certain, because a stale quote or a mismatched pair can read as a large gap that was never really there. None of those are in these numbers. A separate caveat rides on many of the pairs that are counted: one venue often states how it settles a cancelled game while the other stays silent, so the two could pay out differently in that tail case. We disclose that per market, and it is why a pair can sit in manual review even when the outcome match itself is certain, and it is worth checking before trading any single one of them. The figures come from one 24-hour window and will move with the slate. What holds from window to window is the order of magnitude, measured in seconds.

The matched feed and the open gaps are on REST, free to read on any plan. The real-time push, where each gap reaches you the instant it opens, is on Basic and Pro. The Quickstart pulls your first feed on a free key.