Polymarket vs Sportsbooks: How Often Do They Disagree?

Sportsbooks set a price and add a margin. On Polymarket and Kalshi, traders set the price in an order book. So how often do the two disagree just before kickoff, and when they do, which one ends up closer to the result?

We checked 13,688 finished events played between 1 October and 10 November 2026. Each one had a closing moneyline from Polymarket, Kalshi or both, next to prices from at least three major sportsbooks.

The short answer: most gaps are small. The typical gap was 1.4 percentage points for Polymarket and 1.2 points for Kalshi. A gap above 5 points happened in 1.7% of Polymarket events and 0.4% of Kalshi events. On games both venues listed, the median difference between Polymarket and Kalshi was about half a point.

When they disagreed by 3 points or more, results on average landed a little closer to the prediction market. The difference is small and only shows up across thousands of games.

How big are the gaps?

The gap is the largest difference on any outcome between the closing prediction market price and the sportsbook consensus, in percentage points. If the books give the home side 60% and Polymarket gives it 62%, the gap is 2 points.

A gap of 1 to 2 points is a small difference in implied probability.

Bigger leagues, smaller gaps

Gaps were smallest in the biggest markets. In the MLB playoffs, NFL, NBA and NHL the median gap was below one point.

Polymarket priced 79 NFL games and was more than 2 points away from the books in three of them. Kalshi priced 74 and never was. Most of the disagreement comes from lower-tier football and from ITF and Challenger tennis, where order books are thin.

Order book depth points the same way. When the thinnest outcome had less than $1,000 available at the best price, the median gap was about 1.5 points on both venues. When every outcome had more than $10,000, it was below 1 point.

Who was closer to the result?

We scored every event with the Brier score, which measures how far a forecast was from what happened. Lower is better. Across the whole sample:

• Polymarket: 0.4471, against 0.4483 for the sportsbook consensus on the same events.
• Kalshi: 0.4352, against 0.4363 for the consensus on the same events.

Both prediction markets scored slightly better, and a bootstrap confidence interval puts the difference below zero for both. The size of it is about a quarter of one percent of the score.

How you remove the bookmaker margin (the de-vig) matters here. Proportional de-vig scales every outcome down by the same factor. The power method takes more margin off underdogs than favourites, which fits the common view that books load more of their margin on longshots. Neither method gives the true probabilities. Against power de-vigged books, the prediction market lead shrinks to about 0.1% of the score: just below zero for Polymarket, and within the margin of error for Kalshi.

The clearer signal comes from the games where the two sides disagreed by at least 3 points. For each event and venue, we took the outcome the prediction market rated furthest above the books, kept it if the gap was 3 points or more, and checked how often it happened. Events listed on both Polymarket and Kalshi count once per venue.

With proportional de-vig, the books said 74.4%, the prediction markets said 78.4%, and the outcome happened 79.9% of the time across 2,316 outcomes. A 95% interval on that last number, resampling by event, runs from about 78% to 82%. The books' 74.4% sits well outside it and the prediction market's 78.4% sits inside it.

With the power method, the outcome the prediction market rated higher was usually the underdog: books 24.6%, prediction markets 28.8%, and it happened 29.6% of the time across 998 outcomes. The direction is the same. At gaps of 5 points or more, the result fell between the two numbers, and the samples (248 and 120 outcomes) are too small to say more.

Polymarket vs Kalshi

On the 5,532 events where both venues had a usable price, the median largest difference on any outcome was 0.5 points. About 1% were more than 2 points apart, and none were more than 5 points apart. Their Brier scores were the same within the margin of error.

A few examples

• Borussia Dortmund vs Real Betis, Champions League, 4 November. The books had Dortmund at 61.5%. Polymarket said 66.2% and Kalshi 65.5%. Dortmund won.
• Daniel Merida vs Daniil Medvedev, Vienna, 28 October. The books had Medvedev at 75.2%. Polymarket and Kalshi both had him near 79.5%. Medvedev won.
• Anna Bondar vs Mia Pohankova, Wuhan, 12 October. Kalshi made Bondar a 52.7% favourite while the books had her at 46.6%. Pohankova won.
• Alexandra Eala vs Diane Parry, Ningbo, 19 October. Polymarket gave Parry 31.5% and the books 27.8%. Kalshi stayed with the books at 27.2%. Eala won.

Individual games go both ways. The pattern above only appears in aggregate.

Methodology

• Data: Odds-API.io historical closing odds and final results for events played 1 October to 10 November 2026. For each event, the prediction market and sportsbook prices come from the same pre-start snapshot, a median of 16 seconds before the scheduled start.
• Sports: American football, basketball, baseball, ice hockey, football (soccer) and tennis. Football uses the three-way market on the regulation-time score. Other sports use the two-way winner market including overtime.
• Sportsbook consensus: the average de-vigged probability across up to 26 major sportsbooks, including DraftKings, FanDuel, BetMGM, Caesars, Bet365, William Hill, Unibet, Betway, Betano and Superbet. Each event needed at least three books. We dropped any single book price more than 12 points away from the other books, which removed fewer than 1 in 500 book prices.
• Prediction market price: the midpoint of the best bid and best ask on each outcome, rescaled to sum to 100%. The ask includes the spread, so the midpoint is the fairer read of what the market thinks. We kept quotes where the widest bid-ask spread on any outcome was above zero and at most 5 points, which left out about 6% of quotes. With every quote included, Polymarket gaps over 5 points rise from 1.7% to 2.1%.
• Results: events with a void, a tie on a two-way market or a tennis retirement were excluded.

Limitations

• Six weeks of data. The sample is large, but it covers part of one season: the MLB playoffs only, and the first weeks of the NBA and NHL seasons.
• One snapshot. We compared closing prices only, so this study cannot say whether gaps opened earlier in the day and then closed.
• The consensus is an average of mainstream sportsbooks. A sharper single reference could score better.
• Prices, not fills. The midpoint is not a price you can trade at, and Polymarket and Kalshi charge fees on some trades, which we ignored.
• Tennis start times are approximate. For more than a quarter of tennis matches the snapshot was taken over an hour before the scheduled start. Both sides share the same timestamp, but for these matches the prices are less of a true close.
• Small differences. The Brier differences are measurable but tiny. Nothing here says either side is mispriced on any given game.

Getting the data

Odds-API.io returns Polymarket and Kalshi prices in the same format as sportsbook odds, with back and lay prices and depth at the best price. The prediction markets guide covers the fields, and our Polymarket and Kalshi odds data post describes the data in more detail.

Closing prices for Polymarket, Kalshi and the sportsbooks in this study are available through the historical odds endpoints. Kalshi history in the API is reliable from October 2026. To grade your own model against closing odds, see Backtest a Betting Model in Python.