Eight AI models frozen on August 23 had the Senate flipping when the September money did not, and the chamber flipped. The Associated Press called it for Democrats on election night, Tuesday, November 3, in a wire story carried by PBS News at 11:46 p.m. ET, with six flips in by then: North Carolina, South Carolina, Texas, Ohio, Kansas and Iowa. Maine followed at 1:02 a.m. and Michigan held at 2:28, so by morning Democrats had flipped seven Republican-held seats, per NPR's tally of AP calls, and held Georgia, Michigan, Minnesota and New Hampshire, the four Democratic seats on this board. Democrats held 47 seats going in and needed a net gain of four; seven flips clear that with room, and the only race on this board still open is Alaska. Kalshi closed its Senate-control market at 11:46 p.m. ET, the same minute as the PBS stamp, and finalized it at 12:46 a.m.: the Democratic contract paid $1, the Republican contract paid zero.
Fourteen of the 15 races on this board are called. The fifteenth is Alaska, where NBC News had the count at 72.4 percent of the expected vote on Wednesday morning with Senator Dan S. Sullivan at 48.5 percent and Mary Peltola at 47.7, too early to call, and if no one clears 50 percent of first choices, the ranked-choice tabulation scheduled for November 18 decides it. That is the one race on this page whose Kalshi contracts still trade.
Our grade: our price-blind panel of eight AI models, run on August 23 and frozen since, stood on the winning side in 11 of the 14 called races and on chamber control, where it had Republicans at 44 percent to keep the Senate against a market trading at 52.5 cents on them. The market favorite this page carried on September 11 won 10 of the 14, lost three, and in Texas had no favorite at all: Ken Paxton and James Talarico sat dead even at 50.5 cents each on September 11, and the panel leaned Talarico by a hair, 52.5 to 47.5, the one called race where the models took a side the market would not. Both numbers treated Texas as a coin flip; Talarico won by 12, five points ahead of Gina Hinojosa's win over Greg Abbott for governor on the same ballot, per NPR. By Election Eve the traders had passed the frozen panel: the Monday-night favorite won 13 of the 14 called races, missing only South Carolina.
Alaska decides who wins the September grade. The eight models could not agree with each other about it, running from 47 to 58 on the Republican, while the Republican traded at 34.5 cents on September 11, 31 on Election Eve and 15 now. Scored over all 15 races, a Peltola win leaves the September market a hair ahead of the panel, 0.188 to 0.189 on the Brier scale explained at the end of this page, and a Sullivan win hands the panel the whole board, 0.186 to 0.208. The ballots that settle it are still being counted.
The Quick Answer
The grade: the panel had the chamber flipping and 11 of 14 called races; the September market had it staying Republican and 10 of 14, with Texas a tie; the Election Eve market had 13 of 14. Democrats won the Senate on the AP's election-night call and finished the night with seven flips. Of the 15 races on this board, 14 are called and settled on Kalshi; Alaska is not, with a ranked-choice count scheduled for November 18 if no one clears 50 percent. The full September 11 board, and every section it fed, is kept below as the record.
Settlement prices, closing times and the Election Eve closes on this page are from Kalshi's market record, fetched November 4, 2026, at 13:06 UTC; the Alaska quote is from a second pull at 13:50 UTC, and NBC's Alaska count had not changed between the two. The panel's numbers were generated August 23 and were never re-scored on this page, so they are the numbers graded here, against the AP calls and Kalshi settlements; no state has certified yet.
Stay ahead of the markets.
Daily insights and expert picks on Kalshi, Polymarket, and what's moving markets.
Free forever. Unsubscribe anytime.
New to these markets? How to bet on 2026 Senate elections on Kalshi covers the mechanics; a price in cents reads directly as a percentage chance, and a settled contract pays $1 or nothing.
The Board, Graded
Fifteen races, in the order the September 11 board ran them. "Called" names the winner and who called it; the counts are NBC News's unofficial tallies as of 8:05 a.m. ET Wednesday, with the share of the expected vote counted, and every state still has to certify. "Election Eve" is the last trade of the Monday, November 2 session on the contract this page carried as the favorite. "Settled" is what that same contract paid.
| Race | Called | Count (NBC, unofficial) | Election Eve | Settled |
|---|---|---|---|---|
| Texas | James Talarico (D), AP, 11:04 p.m. ET | 55.2 to 42.9, 96.2% in | Paxton 27¢ | Paxton $0 |
| Maine | Troy Jackson (D), AP, 1:02 a.m. ET | 53.8 to 46.1, 86% in | Jackson 73¢ | Jackson $1 |
| Michigan | Abdul El-Sayed (D), AP, 2:28 a.m. ET | 52.5 to 45.5, 96.6% in | El-Sayed 79¢ | El-Sayed $1 |
| Ohio (special) | Sherrod Brown (D), AP, 11:10 p.m. ET | 50.0 to 45.2, 98.8% in | Brown 71¢ | Brown $1 |
| Iowa | Josh Turek (D), AP, 11:37 p.m. ET | 50.5 to 46.4, 98.1% in | Hinson 48¢ | Hinson $0 |
| North Carolina | Roy Cooper (D), AP, 8:24 p.m. ET | 58.7 to 39.1, 99% in | Cooper 97.4¢ | Cooper $1 |
| South Carolina | Annie Andrews (D), AP, 10:39 p.m. ET | 51.1 to 46.7, 99% in | Graham 67¢ | Graham $0 |
| Georgia | Jon Ossoff (D), AP, 7:55 p.m. ET | 59.0 to 41.0, 98.7% in | Ossoff 96.5¢ | Ossoff $1 |
| Nebraska | Pete Ricketts (R), NBC News projection | 50.3 to Osborn 43.7, 90.7% in | Ricketts 67¢ | Ricketts $1 |
| Alaska | Not called. NBC: too early to call | Sullivan 48.5, Peltola 47.7, 72.4% in | Democrat 71¢ | Still trading |
| Kansas | Adam Hamilton (D), AP, 11:25 p.m. ET | 50.9 to 46.1, 98.5% in | Marshall 39¢ | Marshall $0 |
| Minnesota | Peggy Flanagan (D), NBC News projection | 53.7 to 43.6, 98.8% in | Flanagan 90.8¢ | Flanagan $1 |
| Florida (special) | Ashley Moody (R), NBC News projection | 50.9 to 48.0, 99% in | Moody 85¢ | Moody $1 |
| Montana | Kurt Alme (R), NBC News projection | 48.2 to Bodnar 38.1, 83.4% in | Alme 90.8¢ | Alme $1 |
| New Hampshire | Chris Pappas (D), AP (via PBS) and NBC | 56.2 to 41.9, 90.1% in | Pappas 94.4¢ | Pappas $1 |
| US Senate control | Democratic Party, AP, 11:46 p.m. ET | 7 flips, Alaska open | Republican 25¢ | Republican $0 |
The AP times are when NPR published each call (control: the PBS stamp). On the nine races with an AP time, Kalshi closed the market within about ten minutes of NPR's stamp (Texas at 11:05 p.m. ET, Ohio at 11:14, Kansas at 11:28, Iowa at 11:38, Maine at 1:04 a.m., Michigan at 2:34); Nebraska closed at 3:38 a.m., and all 14 were finalized between 8:28 p.m. ET Tuesday (Georgia) and 4:08 a.m. Wednesday (Nebraska), the control contract at 12:46 a.m. Nebraska, Minnesota, Florida and Montana are shown on NBC's projection because no NPR story on them is in our record; the AP's own Senate-control wire names Pappas's New Hampshire win.
Read the Election Eve column against the Settled column and the night gets simpler than the September board made it look. The three races both the September 11 market and the panel missed were Iowa, South Carolina and Kansas, every one a Republican-held seat in a state Trump carried. Four names in this page's September 11 favorite column lost: three favorites, plus Paxton, listed at a tie. The market had walked away from three of them by the night before the vote. Paxton closed Monday at 27 cents against Talarico's 74, Hinson at 48 against Turek's 53 and Marshall at 39 against Hamilton's 62, so the Democrat was the favorite in all three when the polls opened. South Carolina is the one the price never crossed. Darline Graham closed Election Eve at 67 cents, down from the 86.5 this page carried in September, and Annie Andrews won by 4.4 points on NBC's unofficial count in a state Trump carried by nearly 18 in 2024, the first Democrat elected to the Senate from South Carolina since 1998. The panel, frozen since August, got no such chance to walk anything back, which is the point of grading it.
Kalshi's rule for these contracts says each market resolves when the winner is sworn in, which for the regular seats means January. That is not what happened. Each market's secondary rule allows accelerated determination once a consensus of media organizations projects the winner, and the exchange used it on every called race, closing each one the same night and finalizing it by Wednesday morning's pull, so the September edition of this page, which said these contracts settle on who is sworn in, was wrong in practice and is corrected here. Alaska settles the same way once outlets project a winner, most likely after the November 18 tabulation.
Alaska is what is left to trade. The Republican contract is 13 cents bid, 15 offered, last trade 15; the Democratic contract is 85 bid, 86 offered, last 85. That is the market's read of a count in which Sullivan led the first choices 101,285 to 100,311 after midnight Alaska time, 974 votes, with 90 percent of precincts reporting. One reason a price can lean Peltola while Sullivan leads the first choices is in the Anchorage Daily News's count story: many of the precincts still out are rural communities expected to favor her. If no one clears 50 percent, the November 18 ranked-choice count redistributes Gerald Heikes's 2.2 percent (a Republican) and Dan J. Sullivan's 1.7 percent by their voters' next choices, and certification is set for November 25, per the ADN. The two tallies do not line up and neither is official: the ADN's 90 percent counts precincts reporting, while NBC's Wednesday-morning count, 100,751 to 99,074 with 72.4 percent of the expected vote in, is smaller and gives Sullivan a wider lead of 1,677 votes, with roughly 80,000 ballots still out. None of that is a call. NBC News lists the race as too early to call, no call is in our record, and this page will not call it from a price; the 85 cents is what traders think the count will say, and the grade on the panel's Alaska number waits for the state. Our board on the two Dan Sullivans on Alaska's Senate ballot carries the race on its own.
The Board As It Stood On September 11
This is the board the grade above was scored against, unchanged, in the order it ran on September 11. The panel's numbers on this page are estimates, not predictions of fact and not financial advice. A seat was here when at least one of the 12 public handicappers who rate Senate races called it competitive. "Market favorite" was the higher-priced of the two party contracts; "panel" is what our eight models gave that same candidate.
| Race | Market favorite | Kalshi | Panel | Gap | Volume |
|---|---|---|---|---|---|
| Texas | Ken Paxton (R), tied | 50.5¢ | 47.5% | −3.0 | $11.35M |
| Maine | Troy Jackson (D) | 66.5¢ | 65.5% | −1.0 | $5.32M |
| Michigan | Abdul El-Sayed (D) | 62.5¢ | 58% | −4.5 | $3.32M |
| Ohio (special) | Sherrod Brown (D) | 52.5¢ | 62% | +9.5 | $3.08M |
| Iowa | Ashley Hinson (R) | 65.5¢ | 55% | −10.5 | $2.53M |
| North Carolina | Roy Cooper (D) | 91.0¢ | 89% | −2.0 | $2.02M |
| South Carolina | Darline Graham (R) | 86.5¢ | 83% | −3.5 | $1.27M |
| Georgia | Jon Ossoff (D) | 92.7¢ | 82% | −10.7 | $1.24M |
| Nebraska | Pete Ricketts (R) | 70.5¢ | 68% | −2.5 | $0.86M |
| Alaska | Democratic party | 65.0¢ | 48% | −17.0 | $0.85M |
| Kansas | Roger Marshall (R) | 79.5¢ | 82% | +2.5 | $0.78M |
| Minnesota | Peggy Flanagan (D) | 90.2¢ | 90% | −0.2 | $0.57M |
| Florida (special) | Ashley Moody (R) | 91.3¢ | 86.5% | −4.8 | $0.56M |
| Montana | Kurt Alme (R) | 91.8¢ | 81% | −10.8 | $0.51M |
| New Hampshire | Chris Pappas (D) | 83.5¢ | 86.5% | +3.0 | $0.39M |
| US Senate control | Republican Party | 52.5¢ | 44% | −8.5 | $8.11M |
Prices in that table are midpoints as of September 11, 2026, 09:40 UTC, and the last column is as the page carried it at that stamp. The illustration at the top of this page shows the September 4 board.
Every other race board we score, Senate and otherwise, lives on the Kalshi picks hub.
Everything from here down to the closing section, "What The Board Got Right, And What It Missed," is the page as it stood on September 11, kept as the record of what it said before the vote. Its present tense belongs to that date; that closing section and the grade above are the only parts of this page written after the polls closed.
What Moved Between September 4 And September 11
This page's previous board was stamped September 4. Most of the moves since came on or after September 8, when New Hampshire voted; Iowa's started September 4, the day after Emerson's follow-up. Eight of the 16 lines are within a cent of where they were. These are the eight that are not, plus Texas, New Hampshire and the control contract, because the week's story runs through them.
| Contract | Sept 4 | Now | Move |
|---|---|---|---|
| Kansas, Roger Marshall (R) | 83.5¢ | 79.5¢ | −4.0 |
| Nebraska, Dan Osborn (I) | 29.5¢ | 26.5¢ | −3.0 |
| Alaska, Republican | 36.5¢ | 34.5¢ | −2.0 |
| Iowa, Ashley Hinson (R) | 63.5¢ | 65.5¢ | +2.0 |
| Maine, Troy Jackson (D) | 68.5¢ | 66.5¢ | −2.0 |
| Ohio, Jon Husted (R) | 46.5¢ | 48.5¢ | +2.0 |
| Minnesota, Peggy Flanagan (D) | 92.1¢ | 90.2¢ | −1.9 |
| Montana, Kurt Alme (R) | 90.6¢ | 91.8¢ | +1.2 |
| Texas, Ken Paxton (R) | 51.5¢ | 50.5¢ | −1.0 |
| New Hampshire, Democrat, now Chris Pappas | 82.5¢ | 83.5¢ | +1.0 |
| US Senate control, Republican | 52.5¢ | 52.5¢ | 0.0 |
The biggest move is the one with no poll behind it. Kansas opened September 8 at 82 cents on Roger Marshall and printed as low as 78 before closing at 80. It has closed 79 to 80 since, with one dip to 78 on September 10. Adam Hamilton's contract went from 16 to about 21 over the same days, and the largest print of the move was 33,070 contracts at 21 on September 10. Nothing new is on the public polling record: the newest surveys are still the two Democratic-sponsored August polls that had the race inside two points, and the Silver Bulletin average, updated September 9, has Marshall up 3.2. The result is the only one of the six long shots at the bottom of this page where the panel likes the favorite more than the market does. Marshall's 82 from the panel now sits above his 79.5-cent price, so Kansas leaves the long-shot pattern described further down.
New Hampshire did what the schedule said it would. Chris Pappas won the Democratic nomination 65 to 33 over Karishma Manzur and John E. Sununu the Republican one 68 to 27 over Scott Brown, per NBC News's count with 99% reporting, and Kalshi's contracts now carry their names. Pappas is 83.5 cents, one cent above the party line he replaced; Sununu is 16.5. The panel's stored 86.5 stays on the board until the first named-candidate run, which this event was the trigger for.
Maine got its first neutral September poll and moved two cents. CNN's survey with SSRS, taken August 31 to September 6 among 880 likely voters, has Jackson 48 and Collins 45, inside its 3.7-point margin, as the Portland Press Herald reported on September 9. That day Collins's contract traded as high as 38, with the session's largest print 14,445 contracts at 34, and closed at 33. She is 33.5 now, one point from the panel's 34.5.
September 9 was also the day the big money showed up. Prints of 125,000 contracts, the largest on the board this week, landed in Texas on Paxton at 53 cents at 9:16 a.m. Eastern, in Maine on Jackson at 66 at 1:50 p.m., in Ohio on Brown at 50 at 5:39 p.m., and in Iowa on Josh Turek at 36 at 8:52 p.m. The same size hitting four different races inside twelve hours reads like one participant sizing across the map rather than four separate views, though that is an inference from the price history, not a fact it states; at 53 cents the Texas print alone was about $66,000. Texas traded 49 to 54 in that one session on 319,000 Paxton contracts before settling at 51; Brown dipped to 50 and was back at 53 the next day. The day before, David Maxwell, the former director of law enforcement at Paxton's own office and one of the whistleblower plaintiffs in the $6.6 million case described below, endorsed Talarico, per the Texas Tribune.
Two quieter moves. Minnesota's Republican, Michele Tafoya, went from 9 cents to 12 on September 9 after a TIPP poll sponsored by the League of American Workers, a conservative group, put Flanagan up only 44 to 42 among 1,201 likely voters surveyed August 31 to September 2, as Newsweek reported; she is 10.5 now and Flanagan 90.2. In Nebraska, Dan Osborn slid from 30 to 25 on September 8 with a 17,382-contract print on Ricketts at 71 and no new poll since the Democratic-sponsored August tie; he has recovered to 26.5.
Alaska's Republican contract eased to 34.5 on thin volume, and the poll this page is waiting for, a neutral read from a firm other than Alaska Survey Research, still has not arrived. The control contract did not move at all: every close since the last stamp landed at 52 or 53 cents on the Republican side, which is the same thing the page said a week ago.
Every Seat's Number
Eight models, one pass each across the whole map, then a second round in which each read the other seven's anonymized reasoning and could revise. The numbers are each model's final estimate of the Republican's chance in that race, and of Republican Senate control.
| Model | Alaska | Ohio | Texas | Iowa | Maine | Michigan | Georgia | N. Carolina | Control |
|---|---|---|---|---|---|---|---|---|---|
| Claude Fable | 47 | 38 | 47 | 55 | 32 | 42 | 18 | 11 | 44 |
| Claude Opus | 52 | 38 | 47 | 56 | 35 | 42 | 17 | 11 | 46 |
| Claude Sonnet | 56 | 36 | 51 | 54 | 38 | 47 | 15 | 10 | 37 |
| GPT | 47 | 38 | 50 | 55 | 35 | 42 | 20 | 12 | 44 |
| Gemini | 58 | 44 | 47 | 53 | 33 | 42 | 20 | 10 | 47 |
| GLM | 50 | 42 | 47 | 55 | 33 | 40 | 20 | 10 | 42 |
| Kimi | 52 | 36 | 50 | 55 | 34 | 40 | 16 | 12 | 47 |
| DeepSeek | 55 | 40 | 48 | 52 | 35 | 42 | 18 | 12 | 42 |
Every seat is graded against real market settlements, recomputed daily on the public scoreboard. The headline number for each race is the seat median, which our own accuracy audit says tracks the market most closely.
Look down the Alaska column and then down any other. Alaska runs from 47 to 58, straddling a coin flip in both directions. North Carolina, by contrast, runs from 10 to 12. Texas, the race most models named as the truest toss-up on the map, keeps all eight numbers inside a four-point band. The panel calls Texas a coin flip and agrees about it; Alaska is the race it argues about.
More boards from the same panel: Texas Paxton 50.5¢ on September 11, settled $0 · Ohio Brown 52.5¢, settled $1 · Maine Jackson 66.5¢, settled $1.
Alaska: The One Race The Panel Could Not Agree On
Incumbent Republican Dan Sullivan finished second in the state's August 18 top-four primary, behind Democratic former congresswoman Mary Peltola. The count, as NPR carried it, had Peltola at 49.5% and Sullivan at 41.4%. Peltola is the last Democrat to win any statewide election in Alaska, and the general election is decided by ranked-choice voting, so second preferences settle the seat.
Traders read that result the obvious way: the Democratic contract trades at 65 cents on Kalshi and the Republican one at 34.5. The panel's middle number on the Republican is 52. The reason is a single line in the data card, and it is the most useful thing on this page.
Four of the six public general-election polls in the panel's data card came from one firm, Alaska Survey Research. In the graded archive of pollster accuracy the panel was given, that firm carries an average miss of 6.52 points with a documented lean toward Democratic candidates. The only unrated, non-partisan poll in the set, from The New York Times and Siena College in late June, had Sullivan ahead by two. Every model saw the same facts and weighted them differently, and that is the entire spread.
"Alaska (52) is where I depart from the group: four of six polls are Alaska Survey Research, which the card grades at a 6.52-point average miss, D-leaning." — Claude Opus
"Alaska: Peltola leads 5 of 6, but ASR carries a documented 6.5-pt D lean, Sullivan has climbed 43 to 49 in that series, and the only neutral-unrated poll (NYT/Siena) had Sullivan +2; 47." — Claude Fable
Opus at 52 and Fable at 47 are looking at one identical page of evidence and disagreeing about how much to discount one pollster. Gemini, which moved further on this race than any model moved on any race, ended at 58. If the house effect is real, the panel's numbers say the 34.5-cent Republican contract is the most underpriced line on this board. If it has faded since that archive closed, the market is right and Alaska is the panel's worst call of the cycle. That is an open question, which is why it is the one worth reading about.
One poll has landed since the panel ran, and it came from the same firm. Its September 2 read, relayed by Political Wire, has Peltola ahead 51 to 49, inside the margin of error, so it settles nothing on either side. The Alaska ballot has its own complication no poll captures cleanly: our board on the two Dan Sullivans on Alaska's Senate ballot covers what happens when a challenger shares the incumbent's name.
Ohio: The Panel Called It Early, And The Market Had Drifted Its Way
At the panel's August 23 read, Ohio was the tightest market on this page. It has since drifted: appointed Republican Jon Husted trades at 48.5 cents against former senator Sherrod Brown at 52.5, and Husted took back two cents of that drift this week, in the race to complete the term JD Vance gave up to become vice president. Brown lost a different Ohio seat, to Bernie Moreno, in 2024.
The panel is not split on which way it goes. All eight models put Husted below 50, for a median of 38.
Nothing we can source explains the market's drift toward them. The only Ohio poll published since the panel ran, an August 28 Abacus Data survey that the Pollsmax aggregator labels Republican-affiliated, had Brown 52, Husted 46, and it came out after the price had already moved.
The models' case is the polling. Five of the six most recent public surveys have Brown ahead, including two conducted jointly by a Democratic and a Republican firm, the shape of poll least likely to carry a house lean. The only one with Husted in front, from The New York Times and Siena in late June, is the oldest of them.
Ohio's partisan lean is R+5, which is a hard backdrop for a 62% Democratic chance. The models gave it anyway, on the strength of a national environment the card measured at a 6.3-point Democratic generic-ballot lead and presidential approval 21.7 points under water. The full Ohio board has the price history behind the move.
Texas: The Biggest Market, And A Coin Flip (As Written September 11)
Texas carries more money than any other race on the exchange, and it is the race the panel most often named as a true toss-up. Most of the models pointed at Texas when asked which seat was the real coin flip, and their numbers run only from 47 to 51.
The contest exists in its current form because Ken Paxton beat four-term incumbent John Cornyn in the May 26 Republican runoff, after a late endorsement from President Trump, as NPR reported at the time. Democratic state representative James Talarico is the opponent.
Paxton's record is documented and unusually long. He was indicted on felony securities-fraud charges in 2015 and resolved the case through a pretrial agreement involving community service, legal-ethics courses and roughly $300,000 in restitution, according to ABC News. The Texas House impeached him in 2023 on articles alleging preferential treatment for a donor, misapplication of public resources, false statements against whistleblowers and obstruction of justice; the Texas Senate acquitted him on every article that September. The whistleblower litigation brought by his own former senior aides cost Texas taxpayers $6.6 million in settlements, per PBS NewsHour.
In June, Dan Cogdell, an attorney who worked on Paxton's own impeachment defense team, endorsed Talarico, who has built his campaign around that record, as the Texas Tribune has reported. Every allegation above is stated as the charging documents and the reporting state it; Paxton was acquitted, and none of it is offered here as anything but the public record the market has had to price.
The market has priced all of it as a tie, literally: Paxton and Talarico both sit at 50.5 cents in a state whose partisan lean is R+6.
The contract dipped to 49 on August 25, the day after the University of Texas/Texas Politics Project put Talarico ahead 42 to 39, and climbed back within two sessions.
The polling is a tie as well. Emerson College, the only graded-neutral pollster in the Texas set, had Paxton narrowly ahead in mid-August, and The New York Times and Siena had it 47-47 in June. The newest survey, Overton Insights for the Texas Public Policy Foundation, taken August 24 to 26, has Talarico at 44 and Paxton at 43.6.
"Texas — R+6 fundamentals and Emerson Paxton +1 cancel a broader polling record narrowly favoring Talarico." — GPT
The panel's read is that the state's lean and the candidate's record cancel almost exactly, and that the market gives the lean slightly too much credit. On a market this deep, three cents is a rounding argument between two views of the same tie. Our full Texas board walks the polling series in detail.
Iowa, Maine And Michigan: Three Tossups, Three Different Arguments
All 12 handicappers call each of these three competitive. The panel lands close to the market on two of them, and Iowa is the exception, for a reason the polling explains.
Iowa is an open seat, with Joni Ernst retiring, between congresswoman Ashley Hinson and Democratic state representative Josh Turek. Hinson trades at 65.5 cents on Kalshi; the panel's median is 55, with all eight models inside a four-point band.
Emerson had Hinson narrowly ahead in early August, and its follow-up, published September 3, has her up five, 50 to 45. Earlier summer polls had Turek ahead, and the Cook Political Report moved the race to Toss Up on August 20, yet the price has moved Hinson's way since. The market has moved on a poll the panel never saw, which is exactly the kind of gap the grading exists to settle.
Maine is where the market, the raters and the panel agree most tightly. Susan Collins, five terms in and chairing Appropriations, trades at 33.5 cents against former state Senate president Troy Jackson at 66.5.
Eleven of the 12 raters call it a tossup or better for Democrats, and Jackson leads most head-to-head polls taken since July. The panel puts Collins at 34.5, one point from the market. The raters, the money and the models all agree a long-serving Republican in a state that leans Democratic is an underdog. Maine votes by ranked choice, which our Maine board covers.
Michigan is the only race the ratings put at a dead-even partisan lean, an open seat with Gary Peters retiring, and it produced the strangest polling picture on the map. Republican-aligned firms have shown Democrat Abdul El-Sayed ahead by as much as seven; a joint Democratic-Republican poll had Republican Mike Rogers up four. Gemini flagged exactly that when asked where it was most likely to be wrong: house effects running backwards suggest unstable numbers rather than a real lead either way.
The newest read, an EPIC-MRA poll taken August 22 to 28, has El-Sayed up 47 to 43. The market pays 36.5 cents on Rogers and the panel says 42; the Michigan board has each seat's reasoning.
Georgia And North Carolina: Where 93 Cents Looks Expensive
These are the two seats the market treats as nearly finished, and the panel agrees with one of them.
North Carolina is an open seat with Thom Tillis retiring, former governor Roy Cooper against former RNC chair Michael Whatley. Cooper trades at 91 cents and the panel puts him at 89, one of the tightest agreements on the board and the only one with no poll arguing the other way. Every public head-to-head poll since April has Cooper ahead, the newest being High Point University's August 18 read at 50 to 45. There is no argument here, and the price reflects that.
Georgia is different. Jon Ossoff trades at 92.7 cents against congressman Mike Collins, 93 at the offer; the panel puts him at 82, the third-largest gap on the board. Georgia leans R+1, the only seat Democrats are defending in a state that leans Republican at all, and a race without a majority winner goes to a December 1 runoff.
The polling is strong for Ossoff: the Pollsmax average has him up seven and a half across ten polls, and the newest, InsiderAdvantage on August 17, has him 50 to 43.
But five of the 12 raters still class the seat as merely leaning, and the panel prices leaning like leaning. The market prices it closer to done, and a Democratic loss here would turn the party's path from a four-seat gain into a five-seat one. Our Georgia board carries the runoff mechanics in full.
The Long Shots The Board Still Prices
Six races at the quieter end of the battleground lists still carry real money, and they share one shape: the panel sits under the market on five of them. Kansas, as the section under the board explains, crossed over this week.
Nebraska is the most interesting because the challenger is not a Democrat. Independent Dan Osborn trades at 26.5 cents against Republican Pete Ricketts at 70.5, while the Democrat on the ballot, Cindy Burbank, prices at 0.25. Several models named Nebraska as the seat they were most likely to be wrong about, for the same reason: the only head-to-head poll in the panel's record was three months old and sponsored by one side, and the one since, a Democratic-sponsored August survey, has the race tied.
Montana shows the same shape. Republican Kurt Alme trades at 91.8 cents, with independent Seth Bodnar's contract at 6.2 to 7.9. The panel puts Alme at 81, the second-largest gap on the board, in an R+10 open seat.
South Carolina settled its open question. Appointed senator Darline Graham won the August 25 Republican runoff over Ralph Norman, roughly 52 to 48, and the contract that traded as a bare party line at the panel's read now carries her name at 86.5 cents. The panel's 83 was priced against the party line, and the seat gets re-scored now that it has a named candidate. Our runoff board tracked that race to the wire.
Florida's special, Kansas and Minnesota fill out the quiet end. Ashley Moody and Peggy Flanagan are priced as heavy favorites with the panel under each, and Roger Marshall, at 79.5 cents after this week's slide, is now the one favorite among these six the panel rates higher than the market does; the board above has the exact numbers.
Put them side by side and the pattern is the point: a price-blind panel systematically discounts near-certainties, the same habit that produces the Georgia gap on the other side of the map. Our own graded ledger says the habit has not paid so far: where a seat sat five to ten points under a favorite priced at 85 cents or more, the favorite settled Yes on 367 of 368 calls. Most of those were short-dated markets rather than elections, which is why these seats are worth grading on their own.
New Hampshire: The Biggest Move Came From The Smallest Book
The largest slide since the panel ran came from a race this page had not been arguing about. New Hampshire's Democratic contract, 90 cents at the August 23 read, pays 83.5 now, in the open seat Jeanne Shaheen is leaving. The September 8 primary put names on it: Chris Pappas for the Democrats, John E. Sununu for the Republicans, and Kalshi's contracts now carry both.
The sequence is sourced. On August 26 the University of New Hampshire Survey Center released a Granite State Poll that put Republican John E. Sununu at 45 percent and Democrat Chris Pappas at 43, inside the margin of error, a result UNH itself called a statistical tie and still its first read of the cycle with Sununu in front. The contract had already slipped before the release and kept sliding after it. The same poll has Sununu well ahead for the Republican nomination and Pappas ahead of Karishma Manzur on the Democratic side, with Manzur closing.
What makes it worth watching is the book it happened in. New Hampshire is the thinnest market of the 15, and the panel's stored 86.5 sat under the market at the August 23 read. The slide has now carried the price below the panel, one of two races on this board, with Kansas this week, where the market crossed from above the panel's number to below it. Our New Hampshire board re-scores the seat on its first named-candidate run.
Where The Panel Changed Its Mind
The second round is where this exercise earns its keep. On chamber control, the first round produced estimates from 18 to 64, a spread that would have made the headline number meaningless. After one revision round the same models sat between 37 and 47.
Gemini moved further than anyone, on Alaska, and said why:
"Iowa (R+6) and Alaska (R+6) sit just above the waterline; Hinson leads neutral polls like Emerson, and Sullivan's apparent deficits disappear once Alaska Survey Research's 6.5-point D-bias is corrected." — Gemini
That argument, which began as one model's observation, pulled six of the eight models toward it. Alaska's panel median moved from 44.5 to 52 between rounds, the biggest shift on any race.
Texas went the other way, from disagreement to consensus. In round one the models spread 19 points across Texas; in round two they spread four. Nothing new was fetched between the rounds; the models persuaded each other that the state's lean and the polling record cancel.
Opus held its ground on Alaska and flagged its own exposure rather than softening it:
"Alaska at 52 — I am 5-14 points above six of seven other forecasters because I apply ASR's documented D-lean. If that house effect has shrunk since the graded archive closed, or if ranked-choice second preferences consolidate behind Peltola as they did for Democratic-aligned candidates in prior Alaska cycles, I am wrong." — Claude Opus
What Would Change The Panel's Mind (As Written September 11)
Three checkable events, each with the direction it pushes.
- A non-partisan Alaska poll from a firm other than Alaska Survey Research. The whole Alaska disagreement rests on how much to discount one pollster. One clean read from a graded-neutral firm settles it, and the September 2 read was that pollster again, so it does not count.
- The first named-candidate run in New Hampshire. Kalshi now lists Pappas and Sununu after the September 8 primary, so the next pass prices a matchup rather than a label, and it decides whether the market's 83.5 or the panel's stored 86.5 was closer.
- Post-Labor Day polling in Ohio. The panel's case rests on five of the last six surveys. Two clean September reads with Husted ahead would move the panel back toward the market faster than any other input on this board.
Each of those triggers a re-score on this same URL, with the change logged and the as-of stamp updated.
The Map: Why Four Seats Is The Whole Story
Thirty-five seats are up, 22 of them held by Republicans. Democrats defend the other 13 and need a net gain of four. That defense ratio is why the chamber is in play at all, and the threshold is why in play is different from even.
The chamber is decided inside the ten races that three or more raters call live. Seven are Republican-held: Alaska, Iowa, Maine, Nebraska, North Carolina, Ohio and Texas. Three are Democratic-held: Georgia, Michigan and New Hampshire. Every Democratic loss adds a required pickup from the Republican column.
Opus walked the arithmetic, and its version is representative of how the panel got from race numbers to a control number. Read "R-held losses" as seats Republicans are expected to lose and "D-held pickups" as Democratic seats Republicans are expected to take:
"Control: expected R-held losses total 4.52, expected D-held pickups 0.82, net -3.70 against a 4-seat threshold (50-50 keeps R control)." — Claude Opus
That is the mechanism behind a 44% control number sitting beside race estimates that mostly track the market. The expected net is a hair short of the threshold, which means the chamber turns on whichever tossup breaks last. The environment points one way: a 6.3-point Democratic generic-ballot lead, presidential approval 21.7 points under water, and a base rate in which the president's party lost House seats in 18 of the last 20 postwar midterms.
The map points the other way, and 2018 is why the panel takes it seriously: that November, Democrats flipped 41 House seats and still lost two Senate seats, because they were defending most of the seats up. Same night, same voters, opposite chambers.
The control contract has barely moved while the races under it did: every daily close since August 17 has landed between 52 and 54 cents. The chamber-control board prices that question on its own.
The Bottom Line As Written On September 11
Ten of the 15 races are settled arguments, with the money and the machines at the same answer. Ohio has drifted toward the panel's read, and the panel still thinks the market is behind the polling there. Georgia is priced like a finished race by a market looking at ratings that still say leaning. New Hampshire, the thinnest book, produced the biggest move on a poll that called itself a tie. And Alaska comes down to how much you trust one pollster with a measured lean, which is why eight independent models could not talk each other into agreeing.
The single most price-moving thing this board can experience before November is one clean, neutral survey out of Anchorage.
What The Board Got Right, And What It Missed
The September page ended on one sentence: the single most price-moving thing this board could experience before November was one clean, neutral survey out of Anchorage. Alaska is the race still being counted, so the board's organizing argument is the only one on it that has not been graded.
Everything else has. The panel's chamber call, 44 percent for Republicans against a market at 52.5, was right, and the mechanism it gave in August, Opus's expected net of 3.70 seats against a four-seat threshold, undershot a night that produced seven. The panel's Texas lean, 47.5 on Paxton against a market at an exact tie, was the right side by a hair; it said the state's lean and the candidate's record would nearly cancel, and Talarico won by 12. The Ohio call was right and early. The three misses, Iowa, South Carolina and Kansas, do not separate themselves by state lean, since Nebraska, Florida and Montana look the same on paper and the September market and the models both got those right. What separates them is the price each sat at, and the Brier score says what each miss cost: the squared gap between a probability and the result, averaged over the 14 called races, where zero is perfect and a coin flip on everything scores 0.25.
| Forecast | Brier score, 14 called races | Chamber control |
|---|---|---|
| Panel, August 23 | 0.183 | 0.194 |
| Market, September 11 | 0.192 | 0.276 |
| Market, Election Eve | 0.090 | 0.063 |
The panel beat the September market by a hair on the races and on the chamber, a single outcome scored once, and the Election Eve market beat both by a mile, because it had walked off Texas, Iowa and Kansas while the panel, frozen, walked off nothing. Race by race, the three misses cost this much on the same scale:
| Miss | Panel, August 23 | Market, September 11 |
|---|---|---|
| Iowa | 0.303 | 0.429 |
| South Carolina | 0.689 | 0.748 |
| Kansas | 0.672 | 0.632 |
Kansas is the one miss where the panel paid more than the market, because it was the one long shot where the panel sat above the price. Iowa ran the other way: the panel sat 10.5 points under the market, the right direction, and still had Hinson at 55 on a night she lost.
The re-scores the September page promised, on the first named-candidate runs in New Hampshire and South Carolina and on Ohio's September polling, never ran, so the August 23 numbers are the ones graded here and their errors are the errors of August 23. The Georgia habit the page worried about, discounting near-certainties, cost no calls on this map: every favorite priced at 90 cents or more on September 11 won, Florida by under three points. Between 65 and 87 cents, favorites went three for six. Maine, Nebraska and New Hampshire won; Iowa, Kansas and South Carolina lost, and Graham's 86.5 was the highest price to lose.
Fourteen races and the chamber are settled. The fifteenth, the one eight machines could not agree on, is the one Alaska is still counting, and if no one clears 50 percent of first choices, the November 18 tabulation decides both the seat and the grade. Updated 2026-11-04.
Verdicts generated August 23, 2026, after a revision round in which each model read the others' anonymized reasoning. September 11 prices as of 09:40 UTC that day; settlement data as of November 4, 2026, 13:06 UTC, and the Alaska quote as of 13:50 UTC. Panel: Claude Fable, Claude Opus, Claude Sonnet, GPT, Gemini, GLM, Kimi and DeepSeek, all price-blind. This is a neutral, data-driven analysis of market prices, ratings, polling and historical patterns, and it does not advocate for any party, candidate or outcome.
Prices on this page are Kalshi's book. If you also trade on Polymarket, code OS4 gets new users a $50 trading bonus on a $10 deposit — affiliate link; terms as stated by Polymarket; 18+, availability varies by state.
More on this: 2026 Senate Races: Every Kalshi Board We Track · How To Bet On 2026 Senate Elections: Kalshi Odds · Ohio Senate 2026 Odds: Brown Retakes The Lead At 52 Cents · Michigan Senate Race 2026: Five Polls, One Outlier, Odds That Won't Move · South Carolina Senate Runoff Odds: Graham Vs Norman, Kalshi
