Can Kalshi Predict NFL Upsets Better Than Traditional Sportsbooks?

Michael Thompson

Kalshi vs. sportsbooks is the betting argument of 2026, and the sharpest version of it is about upsets. If a crowd of traders really knows something the oddsmakers don’t, the place it should show up is the Sunday a +400 dog walks out of a road stadium with a win. So we tested it on every NFL game we could match, and the short answer is no: they called upsets about equally well.

Michael Thompson, betting strategy analyst at BettingSite.net
Last updated ✓ Fact-checked
The short version
  • We lined up Kalshi’s price in the last minute before kickoff against sportsbook closing moneylines for 359 NFL games, from the start of the 2025 season through Week 5 of 2026.
  • Underdogs won 124 of them (34.5%). Kalshi had given those winners an average of 36.3%. The books had given them 36.6%. Neither saw the upsets coming any better than the other.
  • Overall accuracy was a dead heat: a Brier score of 0.2151 for Kalshi and 0.2154 for the books, a gap well inside the noise.
  • Kalshi’s tiny edge came from favorites, not underdogs. And which venue “liked” the upset dog more flips depending on how you remove the sportsbook’s cut.
  • The two disagreed on who was favored in just 5 of 359 games. They are mostly the same forecast wearing different clothes.
  • If you want to back an upset, the price mattered more than the forecast: every underdog at sportsbook moneylines lost 2.13%; the same bets on Kalshi after fees lost 3.58%.

Kalshi vs. Sportsbooks: How We Ran the Test

A fair test needs both sides measured at the same moment, on the same games, with the same yardstick. Opinions about which venue is “smarter” usually skip all three. Here’s what we did instead.

Pull every Kalshi NFL game market
Kalshi lists a winner contract for each team in every NFL game. We downloaded all 950 settled markets from its public data feed, then pulled minute-by-minute bid and ask prices for each game.
Freeze the price at kickoff
For each game we took the midpoint between the best bid and best ask in the last minute before scheduled kickoff, averaged across both teams’ contracts. A 37¢ midpoint reads as a 37% chance.
Strip the vig from the sportsbook line
We used the closing moneylines in nflverse’s public game file1, converted both sides to implied probability, and scaled them down so they add to exactly 100%.
Score both against what happened
Every game where both prices existed counted: 271 regular-season games and 10 playoff games from 2025, plus 78 games from 2026 Weeks 1-5. Ties were dropped, including the 27-27 Giants-Commanders game on October 11, 20262.

If converting a contract price into a probability is new to you, our explainer on prediction market prices and odds walks through it. The short version is that cents and percentages are the same number on an exchange, which is exactly what makes this comparison possible.

Brier score
The standard way to grade probability forecasts. For each game, take the probability you gave the team that won, subtract it from 1, and square it. Average that over every game. Lower is better. Calling every game 50-50 scores 0.25, so anything meaningfully under that is real skill.

The Overall Scoreboard Is a Dead Heat

Start with the whole sample before narrowing to upsets, because the headline number frames everything else. Across all 359 games, Kalshi’s kickoff prices scored 0.2151 and the de-vigged sportsbook lines scored 0.2154. That’s a difference in the fourth decimal place.

SampleGamesKalshi BrierBook BrierCloser to result
All games3590.21510.2154Kalshi 198, book 160
2025 regular season2710.21230.2121Kalshi 137, book 133
2025 playoffs100.19240.1985Kalshi 8, book 2
2026 Weeks 1-5780.22780.2289Kalshi 53, book 25

To check whether that gap means anything, we resampled the games 4,000 times. The range of plausible differences ran from Kalshi slightly better to the books slightly better, comfortably straddling zero. In plain English, you can’t tell them apart on accuracy. Log loss, a second grading method that punishes confident misses harder, came out the same way: 0.6172 against 0.6177.

Notice how rarely they even disagreed. They named a different favorite in only five games, and every one of those was priced between -105 and -115 at the books. Kalshi got two of the five right and the books got three.

5 of 359
Games where Kalshi and the books picked different favorites
0.85 pts
Median gap between the two win probabilities
10
Games where they were 3+ points apart

What Happened When the Underdog Won

Now the actual question. We defined an upset as any game the sportsbook underdog won, and there were 124 of them, 34.5% of the sample. If Kalshi had an edge on upsets, its prices should have given those winning dogs noticeably more credit before kickoff.

It didn’t. Kalshi gave the eventual winner an average of 36.3% in those games. The books gave 36.6%. Game by game, the sportsbook line was the more generous of the two to the winning dog 68 times and Kalshi 55 times, with one exact tie. That split is close enough to a coin flip that it isn’t evidence of anything.

Here are the eight upsets priced at +350 or longer. Kalshi was more optimistic on seven of them, by a point or two at most, and the books were noticeably more optimistic on the eighth. In every case both venues had the winner as a heavy underdog.

UpsetMoneylineBook chanceKalshi chance
Panthers 16, at Packers 13 (Nov. 2, 2025)+65012.8%13.5%
Panthers 31, Rams 28 (Nov. 30, 2025)+42518.3%19.0%
Vikings 27, at Lions 24 (Nov. 2, 2025)+41018.8%19.0%
Buccaneers 24, at Cowboys 16 (Oct. 8, 2026)+40019.2%19.5%
Browns 13, Packers 10 (Sept. 21, 2025)+36020.9%22.5%
Dolphins 30, Bills 13 (Nov. 9, 2025)+35021.3%22.0%
Browns 20, at Bengals 18 (Jan. 4, 2026)+35021.3%21.5%
Cardinals 26, at Chargers 14 (Sept. 13, 2026)+35021.3%19.0%

Half a point here and there is not foresight. A trader who’d bought the Panthers at Lambeau on Kalshi was paying for a 13.5% shot. The sportsbook bettor was paying for a 12.8% shot. Both were buying a longshot, and both happened to cash.

Neither venue saw the upsets coming. They just disagreed, by a point or so, about how unlikely they were.

The twist: the answer depends on how you remove the vig

A sportsbook moneyline has the house margin baked in, so you have to take it out before comparing it to an exchange price. The simplest method, which we used above, shrinks both sides by the same proportion. Two other common methods assume more of the margin sits on the underdog, so they hand the dog a little less.

Switch methods and the upset result flips. Under the power method, Kalshi gave the winning dog more credit in 84 of the 124 upsets. Under the Shin method, Kalshi led 75 to 49. Kalshi’s overall Brier score was a hair better than all three versions of the book line, but always by a margin too small to matter.

📊 Why this matters for you
Whenever someone claims an exchange “had the upset” and the books didn’t, ask how they turned the moneyline into a probability. A one-point difference in method is the same size as the entire gap between the venues.

Our guide to the break-even win rate shows the raw conversion from American odds before any vig comes out. It’s the starting point for every number on this page.

Where Kalshi Actually Beat the Books

Kalshi was closer to the final result in 198 games to the books’ 160, which is borderline significant on its own. So where did that come from, if not upsets? The answer is the other 235 games, the ones the favorite won.

In those, Kalshi gave the winning favorite more probability 143 times and the books did 92 times. That pattern is real and fits the price data. On average Kalshi priced the underdog at 33.1% and the books at 33.4%. A market that leans slightly harder toward favorites looks smarter in the roughly two games out of three that favorites win.

The catch is that underdogs actually won 34.5% of the time in this sample, a little more than either venue expected. So Kalshi’s lean helped it on favorites and cost it on upsets, and the two effects nearly canceled out.

The 2026 sample leans Kalshi

The 2026 slice is the most interesting part of the data. Through five weeks, Kalshi was closer to the result in 53 of 78 games, a split unlikely to be pure luck. Its 2025 regular season was essentially even, 137 to 133.

One plausible reason is depth. The median NFL game drew about 16.0 million Kalshi contracts in 2026 against 13.9 million in 2025, and the gap between bid and ask at kickoff was a single cent in every 2026 game. Exchange volume as a whole has kept climbing: DeFi Rate counted $19.03 billion in Kalshi contract volume for the week of October 5-11, 2026, the highest weekly total in its six-week review3.

Still, 78 games is five weeks of football, and the Brier gap in 2026 (0.2278 to 0.2289) is small. Treat it as something to watch over the rest of the season rather than a verdict.

The last two hours do real work

We also froze Kalshi’s price two hours before kickoff. At that point it scored 0.2160, worse than both its own kickoff price and the books’ closing line. Two hours out, the books were the more generous venue to the winning dog in 82 of the 124 upsets.

That’s a fairness caveat worth stating plainly. Our Kalshi number is the final minute before kickoff, while the sportsbook number is a closing snapshot. Some late news, like inactive lists, can reach the exchange price after the book figure was recorded. That gives Kalshi a slight timing advantage in this test, not a handicap.

The biggest disagreements point the same way. In the 10 games where the two were three or more points apart, the team Kalshi leaned toward won seven. It’s a tiny sample, but it’s what you’d expect if the exchange is absorbing the last bit of news. For the structural reasons an exchange price moves differently from a book’s, see our breakdown of an exchange vs a sportsbook.

Were Underdogs Priced Right Anywhere?

Accuracy scores blend everything together, so it helps to look at calibration: when a venue said a dog had a 25% chance, did dogs at that price win about 25% of the time? We grouped every underdog by each venue’s own price.

Dog’s priceKalshi: priced / wonBook: priced / won
Under 20%15.0% / 10.9%14.8% / 9.1%
20% to 30%25.1% / 27.1%25.0% / 27.4%
30% to 40%35.2% / 40.7%35.2% / 42.6%
40% to 50%44.5% / 45.0%44.2% / 41.8%

The two columns tell nearly the same story. Big underdogs under 20% won a bit less often than priced on both venues. Mid-range dogs in the 30% to 40% band won more often than priced on both. The sample per band is 44 to 122 games, so a few results either way swing these numbers by several points.

What you won’t find in that table is a band where one venue was systematically right and the other systematically wrong. If the books were blind to some kind of upset, Kalshi would show it here. It doesn’t.

What It Costs to Bet an Upset on Each Venue

If both venues forecast upsets equally well, the practical question becomes where an upset bet is cheaper. We placed a flat one-unit bet on every sportsbook underdog in the sample (358 games, leaving out one -110 pick’em) two ways: at the closing moneyline, and by buying the same team on Kalshi at the ask with its taker fee added.

-2.13%
Return backing every underdog at sportsbook moneylines
-3.58%
Same underdogs on Kalshi, buying at the ask plus fees
-5.36%
Every favorite on Kalshi (books: -5.93%)

Underdogs were cheaper at the books and favorites were cheaper on Kalshi. The reason is the fee. Kalshi charges takers 0.07 times the price times one minus the price, so on a 20¢ underdog the fee is about 1.1¢, more than 5% of what you put up. Before fees, the Kalshi underdog bets actually finished slightly ahead, at +0.85%.

At the books, the margin in this sample landed more heavily on favorites. Our page on prediction market fees runs the full fee curve, and it’s the same conclusion from the other direction: under about 15¢, the exchange stops being the cheap option.

⚠️ These are not betting systems
Every number in this section lost money except one pre-fee figure that no one can actually get. The point is the gap between venues, not a strategy. Backing every underdog blindly is still a losing plan anywhere.

One more caveat: our sportsbook figure is a single closing line per game. A bettor comparing several books can often beat it, which is why line shopping matters more for underdogs than almost anything else you can control.

What Other Research Has Found

We’re not the first to try this. A Claremont McKenna senior thesis by Dylan Cotti, submitted in April 2026, compared Kalshi with DraftKings on 285 NFL games from the 2025-26 season, using Brier scores at three points in the week before each game4.

Cotti found Kalshi slightly more accurate, with a gap he described as small and only marginally significant. After he accounted for how each platform spreads its probabilities between favorites and underdogs, the gap largely went away. That lines up with what we found at kickoff: a hair’s edge for Kalshi that depends on method. One difference: his snapshots were taken days before kickoff, and he found both venues got less accurate as the game approached, while our two snapshots in the final hours ran the other way.

Who forecast the NFL better?
Our test, 359 games at kickoffStatistical tie
Our test, 124 upsets onlyStatistical tie
Cotti thesis, 285 gamesKalshi, marginally
After adjusting for methodGap mostly gone

Research on non-sports markets is murkier. A Vanderbilt study of roughly 2,500 markets in the 2024 election, reported by DL News, scored Kalshi at 78% and Polymarket at 67% on its accuracy measure, behind PredictIt5. The lead author said the paper was not really about accuracy; its main finding was that similar markets were priced differently from venue to venue.

Put the three together and a consistent picture forms. On heavily traded NFL games, prediction markets and sportsbooks land in almost the same place. Neither is a crystal ball, and neither is reliably dumber than the other.

How to Use Both When You’re Hunting Upsets

None of this means the two prices are useless to compare. It means the comparison is a pricing tool, not a crystal ball. If you’re looking at an underdog this weekend, these are the questions that the data says are worth asking.

Before you back an underdog on either venue
Do the two prices actually disagree? In our sample the median gap was under one point. A gap that small is noise, not a signal that someone knows something.
Has late news landed? The exchange price kept improving into kickoff. If an injury or inactive report just dropped, check whether the book has moved yet.
What does it cost after fees? Convert the Kalshi ask plus its fee into odds and compare it to the best moneyline you can find. On longshots, the book usually wins.
Is your edge bigger than the gap between venues? If your read on a dog is only a point or two better than the market, the choice of venue can wipe it out.

The Bottom Line

Can Kalshi predict NFL upsets better than traditional sportsbooks? On 359 games from the 2025 season through Week 5 of 2026, no. Kalshi and the books gave winning underdogs almost identical odds before kickoff, they agreed on the favorite in all but five games, and their overall accuracy was statistically tied.

Kalshi’s small overall edge came from favorites and from trading right up to kickoff, and it has looked stronger early in 2026 as volume has grown. That’s worth tracking. For now, the more useful difference between the two is cost: upsets were cheaper to back at the books, favorites on the exchange. For the bigger picture of how the two industries are colliding, read our look at sportsbooks and prediction markets.

Sources

  1. nflverse: nfldata: NFL game results and betting lines Accessed October 2026.
  2. ESPN: NFL Scoreboard, Week 5, 2026 2026-10-12.
  3. DeFi Rate: Kalshi Hits Record $19 Billion Week As Polymarket US Triples Fall Trade Volume 2026-10-12.
  4. Claremont McKenna College: Prediction Markets vs. Sportsbooks: A Comparison of NFL Forecast Accuracy 2026-04.
  5. DL News: Are Polymarket and Kalshi as reliable as they say? Not quite, study warns 2025-12-05.
Trading event contracts? Event contracts are financial instruments regulated by the CFTC, not sports bets placed with a state-licensed sportsbook. Nothing here is financial or betting advice. A contract can settle at zero, so only risk what you can afford to lose. Most venues require you to be 18+, and eligibility, funding, and which contracts are offered vary by venue and by where you live. Availability is subject to active litigation and CFTC rulemaking, so check what is actually offered to you before you trade.
Gambling Problem? Call 1-800-GAMBLER or visit www.ncpgambling.org. Trading contracts on a game carries the same behavioral risk as betting on it, whatever the regulatory label says.
Bet responsibly. Upsets are memorable precisely because they’re rare, and this page shows that no venue reliably sees them coming. Only wager what you can afford to lose, set a deposit or loss limit before you need one, and never chase losses. You must be 21+ and physically located in a state where sports betting is legal.
Gambling Problem? Call 1-800-GAMBLER or visit www.ncpgambling.org. Free, confidential help is available 24/7. More tools and self-exclusion options are on our responsible gambling page.

FAQ

These are the questions we hear most from bettors weighing an exchange price against a sportsbook line on an NFL underdog.

Is Kalshi more accurate than sportsbooks for NFL games?

Barely, and not by enough to matter. Across 359 NFL games from the 2025 season through Week 5 of 2026, Kalshi’s prices in the last minute before kickoff scored a Brier score of 0.2151 against 0.2154 for de-vigged sportsbook closing moneylines. Resampling the games showed the difference is well within the range of chance. A 2026 Claremont McKenna thesis comparing Kalshi and DraftKings reached a similar conclusion.

Does Kalshi see NFL upsets coming before sportsbooks?

Not in our data. Underdogs won 124 of the 359 games. Before kickoff, Kalshi had given those winners an average 36.3% chance and the sportsbooks 36.6%. The books were more generous to the winning underdog in 68 upsets and Kalshi in 55, a split that is not statistically meaningful, and the result flips if you remove the sportsbook margin a different way.

How do you compare a Kalshi price to a sportsbook moneyline?

A Kalshi contract price in cents reads directly as a percentage chance, so a 30-cent contract is a 30% chance. A moneyline has to be converted to implied probability for both teams, and then the sportsbook margin has to be removed so the two sides add to 100%. Only then are the two numbers comparable.

Is it cheaper to bet an NFL underdog on Kalshi or at a sportsbook?

In our sample, the sportsbook was cheaper for underdogs. Backing every sportsbook underdog at closing moneylines returned -2.13%, while buying the same teams on Kalshi at the ask with its taker fee returned -3.58%. Favorites reversed the order: -5.36% on Kalshi against -5.93% at the books. Kalshi’s fee takes a larger share of your stake on low-priced contracts.

Why do Kalshi and sportsbook odds on the same NFL game differ?

Mostly they do not differ by much. The median gap in our 359 games was 0.85 percentage points, and they named different favorites only five times. Small gaps come from how each venue builds in its cut, from timing, since the exchange trades right up to kickoff, and from where each venue’s customers put their money.

  • Michael Thompson

    Michael Thompson is BettingSite.net's betting strategy analyst. He covers the math behind closing-line value, line shopping, hedging and bankroll management, and shows the numbers before the conclusion. Every piece he writes is edited by Paul Wilson, BettingSite.net's editor-in-chief.

Leave a Comment

BettingSite.net Logo

BettingSite.net is your go-to source for unbiased reviews of the best online betting sites. Our mission is to provide comprehensive, reliable information to help you make informed decisions and enhance your betting experience.

Join Our Community


Gambling Help Online Logo
GambleAware Logo
The National Council on Problem Gambling Logo
21+