Polymarket got the 2026 US midterms mostly right, and the places it got wrong tell you more than the places it got right. The crypto-based prediction market, where traders buy and sell yes/no shares on real events, priced Democrats at 94% to win the House and 76% to win the Senate on the afternoon of Tuesday, November 3. By Wednesday, November 4, the networks had projected Democratic control of both chambers. The big call landed.
The misses were specific, and they were about people the market did not see coming. The sharpest was South Carolina, where Annie Andrews, a Democrat and pediatrician, beat Republican Darline Graham 51.1% to 46.7% to become the first Democrat to win a US Senate race in the state since 1998. Polymarket, and almost everyone else, had the seat staying red.
The call that landed
Polymarket's final pre-close numbers (Democrats 94% for the House, 76% for the Senate) sat almost on top of Kalshi's, the US-regulated event-contract exchange that is Polymarket's main rival: 94% and 75%. Both crowds, trading real money on separate platforms, arrived at the same read.
They were right on the shape of the night. By the morning of Wednesday, November 4, CBS and other networks projected that Democrats would control the Senate with at least 54 seats, with 2 races (Alaska and Mississippi) still uncounted, and that the party had also taken the House, though the final seat count was not yet set.
South Carolina, the seat nobody expected to move
The South Carolina result was the night's genuine upset, and its backstory is why. The seat was open at all only because Lindsey Graham, the state's longtime Republican senator, died on July 11, 2026, of an aortic dissection at 71, a month after winning the Republican nomination on June 9. Governor Henry McMaster appointed Graham's sister, Darline Graham, to the seat, and she carried the party's nomination into November.
Andrews won 1,042,362 votes to Graham's 951,165, a 51.1% to 46.7% result with the rest going to third-party candidates. For a state that had not sent a Democrat to the Senate in nearly 30 years, it was exactly the kind of outcome a market built on recent precedent is structurally slow to price. The contracts leaned Republican; the voters did not.
Texas moved too
South Carolina was not the only deep-red Senate seat to flip. In Texas, Democrat James Talarico, a state legislator and seminary student, defeated Attorney General Ken Paxton with roughly 55% of the vote, the first Texas Democrat to win a statewide race since 1994. Together, the results are the clearest signal of where the market and the real electorate parted ways: on the top-line question of which party controls each chamber, the crowd was right; on individual races in states everyone had written off, it trailed the count.
How much money was trading on it
The cycle pulled serious volume. Kalshi reported about $1.33 billion in midterm trading; Polymarket's congressional markets handled roughly $145 million. Higher combined figures, up to about $1.9 billion, circulated from aggregators using a wider scope that folded in every midterm market, not just the congressional ones, so the figures are not measuring the same thing.
Sean Patrick Maloney, who runs the Coalition for Prediction Markets, an industry advocacy group, has pitched these markets as "an unbiased way to forecast events." The midterms are the cleanest test of that pitch in years, and the scorecard is split: unbiased and accurate in the aggregate, blind to the local story in South Carolina.
What a trader actually saw
A price on Polymarket is a probability, not a prediction of certainty. A seat trading at 70% for one party is the market saying that party wins it 7 times out of 10, which also means it loses 3 times out of 10. South Carolina was one of those 3. That is the uncomfortable part of reading these numbers honestly: a market that priced the Democratic Senate at 76% was telling you, in the same breath, that there was close to a 1-in-4 chance the whole top-line call was wrong. It happened not to be.
What is still uncounted
This is a Wednesday-morning scorecard, not a final one. Alaska and Mississippi had not been called. The House majority was projected but not every district was settled. Governors' races and a full race-by-race comparison against the polls and Nate Silver's FLIPR model (Silver's election forecast published at his Silver Bulletin) will have to wait for the count to finish. What can be said now is narrow and real: the market called control of Congress and missed at least 2 seats it had no recent reason to expect.
The lesson of November 3 is not that the market was smart or dumb. It is that a prediction market is only ever as imaginative as its traders, and no one trading South Carolina in October had lived through a Democrat winning it. The aggregate was easy, because the market reads the mood of a whole country well. The upset was hard, because pricing it meant believing that a thing which had not happened in a generation was about to. Annie Andrews is going to the Senate, and the contracts that said she would not are the most interesting data the 2026 midterms produced.







