Updated August 31, 2026 · 11 min read · by Eric Lindquist
Prices as of August 31, 2026, 16:05 UTC, fetched from Kalshi. A price in cents is the market's implied chance: a 51-cent contract is the market saying roughly 51 percent, and it pays $1.00 if it happens.
The Seat Nobody Is Watching
I grew up in South Dakota, so let me start with the least interesting board on the whole map.
Mike Rounds is running for a third term. Kalshi has him at 95 cents. His challenger is Brian Bengs, an independent, at about 5 cents. The Democrat, Julian Beaudion, withdrew, which left Bengs as the only challenger on the ballot. The exchange still quotes Beaudion at a penny. Total volume across all three contracts: about 46,600, lifetime. Not today. Lifetime.
For comparison, one Senate race on that same exchange has traded 10 million.
That is what a decided race looks like on a prediction market, and it is worth looking at first because most of the 2026 map looks like this. Kalshi lists 34 Senate boards, and 24 of them already have a side priced above 90 cents. They are not the story. The story is the other ten, because Democrats need a very specific number of them, and the market has that number sitting on a knife edge.
Free: The Weekly PM Market Brief
The 8-model panel's graded record, the week's biggest market-vs-model gaps, and what's spiking next. One email, Sundays.
Unsubscribe anytime, one click. We never share your email.
What It Actually Takes To Flip The Senate
The Senate is 53 Republicans and 47 Democrats. Thirty-five seats are up in November 2026: 33 regular Class 2 seats plus special elections in Florida and Ohio. Republicans are defending 22 of them, Democrats 13.
That defense split sounds like a Democratic opportunity, and it mostly is not. Most of those 22 Republican seats are in states that have not been competitive in a decade. Wyoming, Idaho, Oklahoma, West Virginia and Tennessee are all trading above 95 cents for the incumbent party.
The number that matters is four. Democrats need a net gain of four seats to reach 51.
Not three. The reason is buried in the contract's own settlement language rather than its title, and it is the single most misunderstood thing about this market. Kalshi settles Senate control on "the party identification of the President pro tempore of the Senate on February 1, 2027." The President pro tempore comes from the majority, and a 50-50 chamber is not a majority for the party without the tiebreaking vice president. So a 50-50 Senate settles this market as a Republican hold. Democrats have to get to 51 outright.
That is a real bar. It is also why the House and the Senate are priced like they are on different planets. On the same exchange, on the same night, Democrats to win the House trades at 84.5 cents. Senate control trades at 48.5 cents for the same party. Same voters, same ballots, opposite chambers.
There is precedent for exactly that split. In 2018, Democrats flipped more than 40 House seats and lost two Senate seats on the same night, because they were defending 26 of the 35 seats that happened to be up. The chamber that is up for election matters more than the national mood.
The Hotbeds
Here are the boards worth watching between now and November, with Kalshi's midpoint price as of August 31.
| Race | Seat held by | Non-Republican side | Price | Contracts traded |
|---|---|---|---|---|
| North Carolina (Open) | R | Roy Cooper | 91.4¢ | 1.72M |
| Maine | R | Troy Jackson | 69.5¢ | 4.51M |
| Alaska | R | Democratic party | 63.5¢ | 0.79M |
| Ohio (Special) | R | Sherrod Brown | 54.5¢ | 2.47M |
| Texas | R | James Talarico | 50.5¢ | 10.01M |
| Iowa (Open) | R | Josh Turek | 40.5¢ | 2.04M |
| Nebraska | R | Dan Osborn (I) | 31.0¢ | 0.66M |
| Michigan (Open) | D | Abdul El-Sayed | 63.5¢ | 2.87M |
| New Hampshire (Open) | D | Democratic party | 83.5¢ | 0.24M |
These are model estimates and live market prices, not predictions of fact and not financial advice. Kalshi event contracts trade on a CFTC-regulated exchange; participation is 18+ and limited to eligible states, as of August 2026. Every race above has both a live bid and a live ask.
Texas is the biggest board on the exchange, and it is priced within half a cent of even. Ken Paxton beat incumbent John Cornyn in the May runoff and drew state representative James Talarico. Texas has a partisan lean of about R+6, and a Republican there is not usually a coin flip. Paxton also carries a documented record: the Texas House impeached Paxton in 2023 and the Texas Senate acquitted him on all 16 articles tried that September, on charges he misused his office to help donor Nate Paul. Eight of his own senior aides had reported him to the FBI in 2020; the Texas Tribune reported that Paul separately employed a woman with whom Paxton acknowledged an extramarital affair. His decade-old securities fraud charges were formally dismissed after he completed a pre-trial diversion agreement involving 100 hours of community service and restitution. None of that is an allegation about the race, and none of it is our characterization. It is the documented record, sitting alongside a price that has an R+6 seat trading at even money. Readers can weigh how much of one explains the other.
Maine is the widest gap between a record and a price. Susan Collins has won five terms, including a 2020 race she trailed in for most of the public polling. Kalshi has her at 30.5 cents against Troy Jackson. A five-term incumbent priced under a third is the most striking number on the board.
Ohio and Alaska are the other two doors. Sherrod Brown lost Ohio's other Senate seat in 2024. He is not running for that one. This is the special election for the seat JD Vance vacated to become vice president, now filled by appointee Jon Husted, and Brown sits at 54.5 cents in it. Alaska's Democratic side is at 63.5 cents in a state that leans R+6.
Michigan is the seat Democrats can lose. El-Sayed at 63.5 cents in an open, even-PVI seat is the one real hole in their defense, and every point of risk there is another pickup they have to find somewhere else.
If you want the full race-by-race panel numbers rather than just the prices, our Senate race board carries all 15 competitive seats. New to how these contracts work, our Kalshi Senate guide covers the mechanics.
Where The Machines And The Money Split
Now the part I actually wanted to look at.
We ran eight AI models on this market without letting any of them see the price. They got a data card instead: the postwar midterm base rates, the current generic ballot, presidential approval, the full handicapper ratings on all 35 seats, and the poll-error record from the Trump era. Then each model read the other seven's reasoning, anonymized, and could revise.
Here is where they landed on Republicans holding the Senate. The blend is the median of the eight models, not an average, so one outlier cannot drag it.
| Model | First estimate | After revision |
|---|---|---|
| Claude Fable | 72% | 72% |
| Claude Opus | 74% | 73% |
| Claude Sonnet | 62% | 66% |
| GPT | 58% | 65% |
| Gemini | 75% | 73% |
| Grok | 64% | 70% |
| Kimi | 70% | 70% |
| GLM | 52% | 70% |
| Median Of The Eight (The Blend) | 67% | 70% |
| Kalshi | 51.5¢ |
Every seat on this panel is graded against real market settlements — records to date: Claude Fable 86% on 1,207 graded calls · Claude Opus 87% on 1,277 graded calls · Claude Sonnet 85% on 1,254 graded calls · GPT 85% on 7,199 graded calls · Gemini 86% on 5,875 graded calls · Grok 84% on 5,024 graded calls · Kimi 84% on 2,897 graded calls · GLM 81% on 2,738 graded calls. Recomputed daily; the full scoreboard is public.
These are model estimates, not predictions of fact and not financial advice. Every number here gets graded in public once the market settles, and you can see the whole ledger on the full graded scoreboard.
More live boards from the same panel: House control 2026 at 84.5¢ for Democrats · the 2026 Senate race board with Texas at 50.5¢ · how to trade Senate markets on Kalshi. Prices fetched August 31, 2026.
So the panel sits 18.5 points away from the price. Before anyone reads that as a signal, here is the honest context: across 3,762 settled markets, when this panel disagrees with the Kalshi price by 10 cents or more, the market has been right about 70 percent of the time, and that rises to 84 percent when the gap reaches 30 cents. Unanimous disagreements like this one, where all eight models land on the same side of the price, are the panel's single worst pocket. We publish the record because it is the record, not because the gap is an opportunity.
What is interesting is why they diverge, because it is not a mystery. It is arithmetic, and you can check it yourself.
Kalshi lists a board for 34 of the 35 seats up: Louisiana has no 2026 contract, and it is a safe Republican hold that would not change the count. Take each remaining board, multiply the non-Republican contract by its price, add those up across the Republican-held seats, then subtract the expected losses on the Democratic-held ones.
Do it counting Democrats only, and you get an expected net gain of 3.67 seats. Democrats need 4.
Do it again counting the two independents on the board, Dan Osborn in Nebraska and Brian Bengs in South Dakota, and you get 4.03.
Both figures run across all 34 boards, not just the nine in the table above. The components: the Republican-held seats carry 4.94 expected losses in total, of which 4.58 go to a Democrat and 0.36 to an independent, against 0.91 expected losses on the Democratic-held side.
That gap between 3.67 and 4.03 is not a rounding artifact. On the market's own numbers, Democrats fall about a third of a seat short on their own, and the two wins that close the gap belong to candidates who are not Democrats.
Go back to the settlement rule and you see why that matters. This contract pays on the party of the President pro tempore, and the President pro tempore comes from the organized majority. An independent who wins and then caucuses with nobody does not produce a Democratic President pro tempore. So 4.03 is the expected number of seats Republicans lose, which is not the same thing as the number Democrats gain for control. Osborn has pledged not to caucus with either party, a pledge his opponents dispute, and one he would be under real pressure to revisit if the chamber came down to him.
None of that makes 48.5 cents a calculation you can do on a napkin. Expected seats and the probability of clearing a threshold are different things: outcomes correlate, seats come in whole numbers, and the price carries liquidity and fees on top. But it does explain the shape. The race boards put Democrats within a fraction of a seat of a majority, from either direction, which is about as close to genuinely even as a market gets.
The panel ran the same count off the handicapper ratings instead of the prices and got roughly +2 to +3. That is the entire disagreement: the money prices these races more favorably for Democrats than Cook, Sabato, Inside Elections and the rest currently rate them. Alaska is the cleanest example. Every rating table calls it a Tossup or Lean R. The market pays 63.5 cents on the Democrat. Maine is the same shape.
November grades which of the two was better calibrated.
Where The Panel Changed Its Mind
The revision round moved six of the eight models, five of them by more than a point, and the reasons are more useful than the numbers.
The biggest move was 18 points:
"I initially over-weighted the generic ballot and under-weighted the specific PVI and ratings of the Republican-held battlegrounds. Live ratings show only North Carolina as a clear Lean D flip; remaining targets are in R+ or tossup terrain, while Michigan is a vulnerable defensive hold."
— GLM, moving from 52% to 70%
The most specific correction was about the tie rule:
"I moved up from 58% to 65% because the other forecasters clarified the arithmetic better than my round-one weighting: a 50-50 Senate resolves Republican, so Democrats effectively need 51 seats, meaning a net +4 rather than merely tying control."
— GPT, moving from 58% to 65%
One model moved partway and then explained exactly why it stopped:
"I did not move further toward the 70-75 cluster because expected value understates variance in a correlated national environment: if the D+6.3 ballot lead is real and holds, several tossups could break together, making a +4 sweep plausible, not remote."
— Claude Sonnet, moving from 62% to 66%
And one of the highest estimates refused to move at all:
"No forecaster surfaced a fact or arithmetic step I had missed; the spread reflects different weightings of the same seat math, not new evidence. I rejected the lower estimates because they weight the national environment as if it maps directly onto Senate seats. 2018 showed a wave year can still lose the out-party Senate ground when the map is red-tilted."
— Claude Fable, holding at 72%
What Would Change The Panel's Mind
Four things, all of them checkable, with the direction each one pushes.
The generic ballot crossing D+8, or falling under D+4. It currently averages D+6.3 across six aggregators as of August 30. Several models named that threshold themselves: past D+8, a correlated sweep of the red-leaning tossups stops being a parlay and starts being a wave, which pushes the number toward Democrats. Under D+4 and the map does the rest for Republicans.
A ratings change in Maine, Alaska or Ohio. The panel counted off the handicapper tables while the market counted off money. If Cook or Sabato move Alaska or Maine from Tossup to Lean D, the gap closes from the panel's side without a single vote being cast.
Presidential approval moving out of the 38 percent band. It sits at 38.0 approve / 58.7 disapprove, net −20.7, across ten aggregators. Approval back above 42 was the specific indicator two models named as the thing that would most reduce expected Republican losses.
Anything that resolves Texas. It is the largest board on the exchange at 10 million contracts and it is priced at 50.5 cents. It is also the seat where a documented record, rather than partisan lean, is doing the pricing. Texas breaking cleanly one way takes the widest piece of uncertainty off the board.
When This Actually Settles
| Election Day | Tuesday, November 3, 2026 |
| Georgia Runoff, If Needed | December 1, 2026 (if nobody clears a majority) |
| Race Contracts Expire | January 4, 2027 |
| Control Contract Expires | February 1, 2027 |
| How It Settles | Party of the President pro tempore on February 1, 2027; may resolve early on a consensus of media calls |
| Market Size | About 7.8 million contracts traded lifetime, roughly 4.2 million still open |
| Prices As Of | August 31, 2026, 16:05 UTC |
This page gets re-scored when the story moves: a ratings change, a generic-ballot shift, or a resolution in any of the races above.
Hottest Prediction Markets Right Now
- 2028 Democratic presidential nominee · $208M traded
- 2027 Pro Football Champion · $76M traded
- Pro Baseball Champion · $64M traded
- 2028 U.S. Presidential Election winner? · $64M traded
- 2028 Republican presidential nominee · $61M traded
- Bitcoin price at the end of 2026 · $34M traded
Every market above links to our full AI model verdict; browse them all on the OddsShopper prediction markets hub, and see how every settled call actually scored on the full graded scoreboard.
- 2026 Senate Races: Every Kalshi Board We Track
- Why Alani Bankhead (D) Will Win the Montana Senate Race
- Why Andy Barr (R) Will Win the Kentucky Senate Race
- Why Cory Booker (D) Will Win the New Jersey Senate Race
- The 72-Year Streak Traders Are Betting Against
The Bottom Line
Democrats need four seats. Kalshi's own race boards put them at 3.67 on their own, and 4.03 once you count two independents who have not agreed to caucus with them. That is the most precise way anyone has of saying "too close to call," and it is why the chamber trades at 48.5 cents rather than anywhere interesting.
Our panel counted the same map off the ratings instead of the money and got two to three, which is why it sits at 70 percent for a Republican hold. Our own graded record says that when we disagree with the price this widely and this unanimously, the price usually wins. Both numbers go on the board, and November grades them.
If you only follow four boards between now and then, follow the four the market says are likeliest to change hands: North Carolina, Maine, Alaska and Ohio. Texas is the fifth, and it is the biggest board on the exchange, so it will make the most noise. Then keep half an eye on Nebraska, because a race almost nobody is watching is carrying about a third of a seat of the majority.
South Dakota will still be sitting at 95 cents, and I will still be checking it.
Hero illustration: OddsShopper, in the house collage style. United States Capitol photo by Beethoven, licensed CC BY-SA 4.0; photos cropped, toned, and composited.
