Updated August 22, 2026 · 13 min read · by Jake Hari. Panel verdicts were generated August 22, price-blind, from the fetched 2028 Senate map, the locked 2024 carryover, presidential-year base rates back to 1980 and the current 2026 midterm environment. Prices as of August 22, 2026.
On Kalshi, the federally regulated exchange where people buy and sell contracts on real-world events, a contract that pays $1 if Democrats win the Senate in 2028 costs 57 cents. A contract that pays $1 if a Democrat wins the White House that same November costs 58 cents.
One cent. Two very different elections, and the market is treating them as very nearly the same bet.
The Quick Answer
Traders currently make Democrats modest favorites to control the Senate seated in January 2029, at 57 cents on the dollar. We asked eight AI models to price the same question without letting any of them see the market. After a round in which each model read the others' reasoning anonymously and could revise, the panel's middle number came out at 51 for Democrats. That is a coin flip, and it sits about six points under the market. The reason the panel gives is a structural one that has nothing to do with who wins the presidency, and it starts with which seats are actually on the ballot.
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.
Why This Is Not a Bet on 2028
The most common mistake on this market is reading it as a referendum on the 2028 election. It is not, and the arithmetic is unforgiving about it.
Senate seats come in three groups, called classes, and only one group faces voters in any given election year. Of the 100 senators seated in January 2029:
- 33 seats are already locked (Class 1). They were filled in the 2024 election and do not come up again until 2030. That group breaks 19 to 14 in favor of Democrats and the two independents who vote with them. None of them is on the ballot in 2026 or 2028, so absent a vacancy or a party switch, that split is fixed.
- 33 seats are decided on November 3, 2026 (Class 2), this year's midterms, which we score seat by seat on the 2026 balance-of-power board.
- Only 34 seats are on the ballot in November 2028 (Class 3). That is the group the models spend most of their reasoning on below.
So Democrats start from 19 and need 32 of the remaining 67 to reach 51. Republicans start from 14 and need 37. There is one shortcut: a 50-50 Senate is broken by the vice president, so whichever party wins the White House in 2028 can run the chamber with 50 instead of 51. That is the real thread connecting this market to the presidential race, and it is worth exactly one seat. Two thirds of this market is settled before a single 2028 ballot is printed, and a big chunk of it is settled this November.
All 34 seats up in 2028 were last on the ballot in 2022, which matters because how narrowly a senator won last time is a usable read on how safe the seat is now. Two of them have appointed senators sitting in them today: Marco Rubio resigned Florida's seat in January 2025 to become Secretary of State, and JD Vance resigned Ohio's to become vice president. Both seats get special elections on November 3, 2026 to fill out the current term, and then return to the regular cycle in 2028.
The Board
| Outcome | Kalshi price | AI panel (median) | Gap |
|---|---|---|---|
| Democratic Senate majority | 57¢ | 51.0% | Market +6.0 |
| Republican Senate majority | 46¢ | 49.0% | Panel +3.0 |
Prices as of August 22, 2026. The panel figure is the blend, meaning the middle number of the eight models rather than their average, so one outlying model cannot drag it. Kalshi quotes the two sides separately, which is why they sum to more than 100 cents; the difference is the spread between what buyers bid and sellers ask.
Every Seat's Number
Each model priced both outcomes in a single pass, without ever seeing the market. Round 1 is the first independent answer. Round 2 is after each model read the other seven's reasoning, anonymized, and was free to revise or hold.
| Model | Round 1 (D) | Round 2 (D) | Moved |
|---|---|---|---|
| DeepSeek | 42.5% | 63.0% | +20.5 |
| Gemini | 65.0% | 54.0% | −11.0 |
| ChatGPT | 56.0% | 53.0% | −3.0 |
| GLM | 48.0% | 53.0% | +5.0 |
| Fable | 57.0% | 49.0% | −8.0 |
| Sonnet | 57.0% | 46.0% | −11.0 |
| Kimi | 55.0% | 44.0% | −11.0 |
| Opus | 44.0% | 42.5% | −1.5 |
| Panel median | 55.5% | 51.0% | −4.5 |
Every model on this panel is graded against real market settlements — records to date: Gemini 86% on 4,423 graded calls · GLM 81% on 2,437 graded calls · Kimi 82% on 2,414 graded calls · DeepSeek 80% on 2,462 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, on the full graded scoreboard.
More live boards from the same panel: 2026 Senate control, Republicans 54¢ · 2026 House control, Democrats 85¢ · 2028 presidential race, Democrats 58¢. Prices fetched August 22, 2026.
The Asymmetry the Panel Kept Finding
Six of the eight models moved toward Republicans after reading each other, and they nearly all moved for the same reason.
The group of seats up in 2028 is usually described as 19 Republican-held and 15 Democratic-held, counting the Florida and Ohio seats currently filled by Republican appointees, which sounds like Republicans have more to defend. Look at which seats are actually close and it points the other way. Sorted by how much of the vote each incumbent won in 2022, the Democratic side of the ledger is where the thin margins live: Nevada at 48.8%, Pennsylvania at 51.2%, Arizona at 51.4%, Georgia at 51.4%, New Hampshire at 53.5%. Two Republican seats sit in the same range on paper, Wisconsin at 50.4% and North Carolina at 50.5%, and two more look close until you read the races behind them. Mike Lee's 53.2% in Utah came against an independent, with no Democrat on the ballot at all. Lisa Murkowski's 53.7% in Alaska came in a ranked-choice count against a fellow Republican. A low winning share in a safe Republican state is not a Democratic opportunity, which is why the panel counted two live Republican targets rather than four.
Sonnet worked the 2026 leg to a modest Democratic gain, which leaves 66 seats settled before 2028 at 32 Republican and 34 Democratic. Republicans would then hold 19 of the 34 seats on the 2028 ballot, and 32 plus 19 is 51. Sonnet put the consequence in one line:
Republicans only need Class 3 to hold serve (net zero) to reach 51; Democrats need a net +2 swing, and the marginal-seat map doesn't clearly hand them that. — Sonnet
Opus, the most Republican-leaning model on the board, built the same point into the reason its number barely moved:
The decisive fact is exposure reversal: Democrats defend five sub-54% seats (NV 48.8, PA 51.2, AZ 51.4, GA 51.4, NH 53.5) against only two genuinely marginal R seats (WI 50.4, NC 50.5). — Opus
There is a second reason the panel would not price a large swing. Presidential years do not move the Senate much. Across all twelve presidential elections from 1980 through 2024, the average change was 3.7 seats in either direction and the median was 3. Only two moved more than four: 1980, when the Reagan landslide flipped 12 seats and handed Republicans the chamber for the first time since 1955, and 2008, when a financial crisis broke weeks before the vote. Several models said plainly that they could not name a comparable condition two years out, so they declined to price one.
Nevada, Pennsylvania, Arizona: The Seats That Decide It
Catherine Cortez Masto won Nevada in 2022 with 48.8% of the vote, the narrowest win of anyone on the 2028 ballot. Hers is the seat almost every model named first when asked what breaks the tie. Pennsylvania's John Fetterman won with 51.2%, Arizona's Mark Kelly with 51.4%, Georgia's Raphael Warnock with 51.4%.
Those four states are also presidential battlegrounds, which is exactly why the models split. If a Democrat wins the White House in 2028, those seats get easier for Democrats to hold. If a Republican wins, one or two of them become live Republican flips, and the chamber can go with them. Kimi described the fork directly:
Nationalization cuts both ways: a Democratic presidential win protects those very swing-state seats and exposes WI/NC; a Republican win flips NV and likely more. — Kimi
That is the argument for why the Senate contract and the presidential contract should be correlated. It is not an argument for why they should cost the same. Opus priced a Democratic presidency at 57 and a Democratic Senate at 42.5, a 14.5-point gap, because winning the White House still leaves the party needing to run the table on four thin seats and then flip one more. The market's gap between the two is one cent.
Three Senators Who Could Change the Answer Without an Election
The record here is not about the map. Senate control can change because someone switches sides, and the three senators whose party loyalty has been most publicly in question all sit in the group up in 2028.
Lisa Murkowski, the Alaska Republican, has said openly that she has considered joining the Senate Democratic caucus, calling it an "interesting hypothetical" and telling Semafor there is "some openness to exploring something different than the status quo." She has also told The Hill she does not see Democrats as much better than Republicans and has real policy disagreements with them, so this is an open question rather than a decision.
John Fetterman has spent two years denying he will leave the Democratic Party, even as talk of a switch has trailed him through his votes for Trump Cabinet nominees, his public criticism of fellow Democrats and his regular Fox News appearances. He has told Spotlight PA he would formally switch only if the party made opposition to Israel part of its official platform, and the Pennsylvania Working Families Party has said it will back a primary challenger against him in 2028.
Mark Kelly's case is the sharpest of the three. In November 2025 the Pentagon opened a review of Kelly, a retired Navy captain, over a video in which he and five other Democratic lawmakers reminded service members of their duty to refuse illegal orders, and said it could recall him to active duty (CNN, NPR). Kelly called the investigation an attempt to chill military dissent and sued Defense Secretary Pete Hegseth. In February 2026 a federal judge granted a preliminary injunction blocking the action, writing that the department had "trampled on Senator Kelly's First Amendment freedoms"; the government's appeal was argued in May 2026.
None of that is priced by a seat map, and every model was given the record. Most of them assigned it a small tail in both directions. Fable was blunt about the size of the effect: a formal switch by either Murkowski or Fetterman would move this market five to eight points on its own, which is about the size of the entire gap between the panel and the price.
Where the Panel Changed Its Mind
The revision round moved this board more than most. DeepSeek reversed itself completely, going from 42.5 to 63 for Democrats after re-weighting the 2026 leg:
A D+6.3 generic ballot and −21.7% approval in a 2026 map that protects most Democratic incumbents produces a D net gain of 3, then a likely Democratic 2028 presidential win delivers the decisive seat in Wisconsin. — DeepSeek
[Editor's note: DeepSeek's revised path counts the Florida and Ohio special elections as part of the 2026 group. Both are 2028-class seats being filled early by special election, so they face voters again in 2028. Opus flagged the same point independently. We publish model reasoning verbatim, including where our own data card disagrees with it.]
Gemini went the other way by eleven points, from 65 down to 54, after conceding the map problem it had underweighted:
Democrats' projected 2026 carryover gains from a hostile midterm environment slightly outweigh their asymmetric 2028 map exposure, with the 50-50 presidential tiebreak favoring the out-party. — Gemini
And Opus barely moved at all, from 44 to 42.5, holding the position the rest of the board drifted toward:
Even a Democratic presidential win barely reaches 51, because Class 3 is the class where Democrats — not Republicans — defend the genuinely marginal seats. — Opus
The pattern is worth naming. In round one the panel's middle number was 55.5, within striking distance of the market's 57. Reading each other pulled it to 51. Whatever the models learned from one another, it pushed them away from the price rather than toward it.
What Would Change the Panel's Mind
Every model was asked what would move its number. Four triggers came back repeatedly, all of them checkable.
- The November 3, 2026 result. This is the big one, and it arrives in roughly ten weeks; traders currently make Republicans slight favorites to hold the Senate this year while giving Democrats the House comfortably. A Democratic net gain of four or more seats in the midterms lowers the 2028 requirement and moves the panel toward Democrats by eight to twelve points. A net gain of zero or one pushes it hard the other way; Opus said that case takes it to Republicans at 65.
- The first real 2028 presidential polling, or a settled ticket. Because Nevada, Pennsylvania, Arizona and Georgia all track the top of the ballot, a clear Republican lead pushes this market toward Republicans sharply, and a clear Democratic lead does the reverse.
- A formal caucus switch. Murkowski joining the Democratic caucus, or Fetterman leaving it, moves the number five to eight points immediately, with no election involved.
- A named 1980-or-2008-scale shock. Several models said they would price outside the low-variance base rate only if someone could name the condition. Absent one, they would not.
Settlement Timeline
| What | When |
|---|---|
| 2026 midterms decide 33 of the 67 undecided seats | November 3, 2026 |
| Florida and Ohio special elections fill two 2028-class seats early | November 3, 2026 |
| 2028 Senate election, 34 seats | November 7, 2028 |
| New Senate seated; control determined | January 3, 2029 |
| Market settles | February 1, 2029 |
This page is re-scored when the story moves. Prices and panel numbers as of August 22, 2026.
One More Thing About the Price
This board is thin. The two 2028 Senate contracts have traded about $3,400 combined since they opened in May 2026. The 2026 Senate market next door has traded roughly $7.5 million over its life, and the two-way market on which party wins the 2028 presidency, the one quoted at 58 cents above, about $536,000. (The separate market on which individual candidate wins is far larger; they are different contracts.)
That context cuts in a specific direction. A 57-cent quote on a book that has barely traded is not the verdict of a large, motivated crowd; it is a handful of orders sitting near the number of the much more liquid market beside it. Which is a decent explanation for why the Senate contract and the presidential contract sit one cent apart, and for why a barely-traded market can drift six points from the panel.
Kalshi is regulated by the CFTC as an exchange for event contracts, and trading is 18+. Contracts on this board settle in cents on the dollar, and a thin book means the price you see may not be the price you get.
Hottest Prediction Markets Right Now
- 2028 Democratic presidential nominee · $188M traded
- 2027 Pro Football Champion · $69M traded
- 2028 U.S. Presidential Election winner? · $61M traded
- 2028 Republican presidential nominee · $59M traded
- Pro Baseball Champion · $58M traded
- Bitcoin price at the end of 2026 · $32M 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.
The Bottom Line
The market says Democrats are favored to hold the Senate in 2028 by about the same margin it gives them in the presidential race. The panel, reading the map without the price, says the Senate is harder than that, because the class on the ballot in 2028 is the one where Democrats defend Nevada, Pennsylvania, Arizona and Georgia while Republicans defend Wisconsin and North Carolina.
Six points is not a scandal. But the honest read is that most of this question gets answered on November 3 of this year, not in 2028, and that a market which has traded $3,400 has not really answered it at all.
