Michigan's Democratic Senate primary was priced as a formality. The last eight public polls of the race had Abdul El-Sayed ahead by 10 to 18 points, the newest of them an Emerson College survey five days out with him up 15, and Kalshi's nominee contract traded at 97.9 cents on primary morning. He won by one point. Overnight the general-election board on the same exchange lurched, recovered, and finished the night within three cents of where it started, which is a strange outcome for a night that produced this much new information. So we re-ran the whole thing: eight AI models, none of them shown a single price, scoring the November race now that the matchup is finally fixed.
New to Kalshi event markets? Start here → how Kalshi's yes/no contracts work.
The Quick Answer
Abdul El-Sayed (Democrat) won the August 4 Democratic primary with 48.5% to Haley Stevens's 47.5%, a margin of about 14,900 votes out of more than 1.5 million cast. He faces Republican Mike Rogers on November 3. Kalshi's general-election market makes El-Sayed a clear favorite at a 62.5-cent mid, with Rogers at 39.5 cents. Our eight-model panel, scoring price-blind after a round in which every model read the others' reasoning, lands at 49.9% El-Sayed and 50.1% Rogers. That is a 12.6-point disagreement: the exchange has a clear favorite, and the panel has a coin flip.
- Market Favorite: El-Sayed, 62.5¢ (down from 65¢ before the polls closed)
- Model Blend: El-Sayed 49.9%, Rogers 50.1%, from all eight seats
- The Gap That Is The Story: the panel prices El-Sayed 12.6 points below the market
- What The Models Keep Pointing At: Rogers led the pre-primary polling average, and the Democrat who polled better against him is the one who lost the primary
- What The Market Keeps Pointing At: a D+6.6 generic ballot, presidential approval 22 points net negative, and a state that sent a Democrat to the Senate by 0.3 points in 2024 with this same Republican on the ballot
The full board, the hour-by-hour record of what the exchange did while the votes were counted, and every model's number are below.
Get the panel's verdict when this settles. We grade every model call in public once a market resolves. Michigan settles November 3, and this page re-scores whenever the story moves. Grab our free updates to get the re-scored board and the settlement grade. No pick, no code, just the receipts.
Prices on this page refresh during the news wave. Every number below carries an as-of stamp, and the whole board was re-fetched at 10:45 a.m. ET on August 5, after the concession and after the exchange finalized its nominee market.
The Board
| Outcome | Party and path | Market | Model blend |
|---|---|---|---|
| Abdul El-Sayed | Democrat; won the August 4 primary by about a point | 62.5¢ | 49.9% |
| Mike Rogers | Republican; unopposed, lost Michigan's other Senate seat by 0.3 points in 2024 | 39.5¢ | 50.1% |
| Haley Stevens | Democrat; lost the primary, so her contract resolves No | 0.5¢ | ~0% |
| Mallory McMorrow | Democrat; withdrew, so her contract resolves No | 0.5¢ | ~0% |
These are model estimates, not predictions of fact and not financial or voting advice.
Every Seat's Number
| Model | El-Sayed, round 1 | El-Sayed, final | Implied Rogers | Movement |
|---|---|---|---|---|
| Claude Fable | 57% | 57% | 43% | held |
| Claude Opus | 56% | 55% | 45% | −1 |
| Claude Sonnet | 47% | 47% | 53% | held |
| GPT (ChatGPT Pro) | 54% | 54% | 46% | held |
| Gemini | 58% | 48% | 52% | −10 |
| GLM | 40% | 45% | 55% | +5 |
| Kimi | 45% | 48% | 52% | +3 |
| DeepSeek | 45% | 45% | 55% | held |
| Blend | 50.3% | 49.9% | 50.1% | spread narrowed 18 → 12 |
| Kalshi Market | 62.5¢ | 39.5¢ |
Every seat on this panel is graded against real market settlements — records to date: Claude Fable 84% on 334 graded calls · Claude Opus 85% on 385 graded calls · Claude Sonnet 83% on 366 graded calls · GPT 83% on 2,013 graded calls · Gemini 85% on 2,062 graded calls · GLM 81% on 2,138 graded calls · Kimi 83% on 2,099 graded calls · DeepSeek 79% on 2,145 graded calls. Recomputed daily; the full scoreboard is public.
These are model estimates, not predictions of fact and not financial or voting advice. The final column is each seat's number after it read the other seven models' anonymized reasoning and was invited to revise or hold. The blend is the simple mean of those eight numbers. Every number in this piece gets graded in public once the market settles, in the full graded Model Verdict scoreboard.
More live boards from the same panel: North Carolina Senate, where Roy Cooper trades at 91¢ · Maine Senate, where Susan Collins sits at 37¢ · the whole 2026 Senate map. Prices fetched August 5, 2026.
What Moved Overnight
This is the part worth studying, because the exchange wrote a record of what traders believed at each stage of the count. All times are Eastern, and all prices are the last trade in that hour, taken from the exchange's hourly candle record. The two largest moves below are the two busiest hours of the night by volume, so they are trades rather than a market maker requoting an empty book.
- 4 P.m., Polls Still Open. El-Sayed 65¢, Rogers 36¢. Both had been parked there for four days.
- 9 P.m., First Returns. El-Sayed ticks up to 66¢, the high of the entire window. The early count matched what the polls and the price had implied.
- 10 P.m. El-Sayed 61¢ on the busiest hour of the night. Stevens's general-election contract, priced at 2¢ that morning, trades more than 19,000 in that single hour.
- 11 P.m., The Low. El-Sayed 54¢, Rogers 40¢, Stevens's general-election contract up to 6¢ from 2¢. The nominee market told the same story more loudly: El-Sayed's contract there fell from 98¢ to 83¢ in that hour while Stevens's climbed from 2¢ to 16¢. For about two hours the exchange was pricing a real chance that Michigan Democrats had picked the other candidate.
- 2 A.m. El-Sayed back to 63¢ as the remaining precincts broke his way.
- 6 A.m. El-Sayed 65¢, all the way back to where he started.
- 10:45 A.m., After The Concession. El-Sayed 62.5¢ mid, Rogers 39.5¢ mid, Stevens bid at zero.
Net for the night: El-Sayed down about 2.5 cents, Rogers up about 3.5. A twelve-cent round trip, and the board finished roughly where it opened. Whatever the market learned from a one-point primary, it did not price much of it into November.
The Market At A Glance
| Venue | Kalshi, a CFTC-regulated event-contract exchange (18+; availability varies by state, as of August 2026) |
| The Market | "Who will win the 2026 Michigan Senate race?" (Kalshi event KXMISENATE-26), one yes/no contract per person. The binding rule reads: "If Abdul El-Sayed wins the 2026 Michigan Senate race in 2026, then the market resolves to Yes." Every other person-contract resolves the same way for its own candidate |
| Settlement | Expiration November 3, 2026. Kalshi's rules add that a candidate who fails to secure the nomination resolves No immediately, and that the market is eligible for accelerated determination once a consensus of media organizations projects a winner |
| The Primary Market, Now Settled | "Michigan Democratic Senate nominee?" (KXSENATEMID-26) is finalized. El-Sayed's contract settled YES; all eight others, Stevens and McMorrow included, settled No |
| Liquidity | Open interest of about 194,000 on the El-Sayed line and 162,000 on Rogers, with roughly 306,000 and 188,000 traded to date. Real depth by this lane's standards, though the quote fields sat null through the night, so every price here comes from the exchange's own hourly candle record |
| Snapshot | Prices as of August 5, 2026, 10:45 a.m. ET. Verdicts generated August 5, 2026 |
These are model estimates, not predictions of fact and not financial or voting advice. Kalshi event contracts trade on a CFTC-regulated exchange; you must be 18 or older and in an eligible state to participate, and this article does not advocate for or against any candidate or party.
What Actually Decides This Race
Six fetched facts define the November board, and they pull hard against each other.
The primary result is a data point about the nominee, not just about the primary. El-Sayed took 48.5% (740,943 votes) to Stevens's 47.5% (726,045), with McMorrow, who had withdrawn but stayed on the ballot, at 4.0% (61,249) out of 1,528,237 ballots cast, per the results carried by NBC News. The count is unofficial pending the state canvass. Stevens conceded and endorsed him Wednesday morning, and McMorrow endorsed him as well. Every model in the panel treated the near-split as live evidence, and they disagreed about which direction it points.
The polling average had the Republican ahead before any of this happened. The three published aggregators, all updated August 2, put Rogers in front: 270toWin at Rogers +4.4, Race to the WH at +2.1, and RealClearPolitics at +0.3. The average line published beneath them reads Rogers 45.3 to El-Sayed 43.1, an edge of 2.2 points, with 11.6% still undecided. The newest individual survey, EPIC-MRA in the field July 24 to 31, had Rogers 46 and El-Sayed 43.
The Democrat who polled better against Rogers is the one who lost. Across the complete public record, Stevens ran roughly two to four points ahead of El-Sayed in the same pollsters' head-to-heads against the same opponent. EPIC-MRA is the cleanest illustration: the same survey, the same week, had Rogers up three on El-Sayed and down two to Stevens. That is the single fact the bearish half of the panel kept returning to.
Michigan Democrats nominated the candidate that public pollsters had running two to four points worse against Mike Rogers, and they did it by fourteen thousand votes. Everything the panel argues about downstream of that is a question of how much it matters.
The national environment points the other way, hard. The generic congressional ballot sits at D+6.6, with all six aggregators clustered between D+6.0 and D+6.9, and presidential approval is 37.7 approve to 59.8 disapprove, a net of −22.1. The president's party has lost House seats in 18 of the last 20 midterms. The two exceptions needed conditions that are nowhere in evidence: 1998 required an impeachment backlash, and 2002 required approval above 60 during a rally-around-the-flag. The same wind blows through every open-seat board this cycle; on Iowa's open-seat Senate race, the same price-blind method still keeps Republican Ashley Hinson favored in spite of it.
Michigan decides these races by fractions. In 2024, running for Michigan's other Senate seat, Elissa Slotkin beat this same Mike Rogers 2,712,686 to 2,693,680, a margin of 19,006 votes, or 0.34 points. Donald Trump carried the state at the top of that same ballot. Michigan splits its ticket and settles statewide contests inside a couple of points, which is why nobody on either side of this disagreement is arguing for a comfortable margin.
Every race rating on the public record predates the primary. Cook Political Report and Sabato's Crystal Ball both call it a toss-up; Decision Desk HQ, Fox News, and The Economist have it Lean D. The newest of those ratings was published July 21, two weeks before Michigan Democrats told everyone something new about their nominee. Rating agencies have not yet had their say on the matchup that actually exists.
Where The Panel Split
All eight seats started from the same place: the midterm base rate and an approval rating 22 points net negative put the Democratic nominee above even in a swing state. What separated them was how much weight to give the polling gap between the two Democrats.
The five seats that landed at or above 48% treated the candidate-specific drag as real but bounded, and already priced into the polls the models were reading.
An open-seat Michigan race with a D+6.6 generic ballot and net −22 presidential approval favors the Democratic nominee slightly despite a real but bounded (about two to four point) candidate-specific El-Sayed drag that pre-primary polls, showing Rogers +2.2 with about 11.6% undecided, already reflect before any consolidation. — Claude Fable (84% on 334 graded calls)
The three seats below 48% weighted the primary result itself as fresh evidence rather than noise. DeepSeek's read was the sharpest version of that case: the counterfactual polls and the one-point win are two independent measurements of the same weakness, and neither has been tested since.
The counterfactual Stevens-vs-Rogers polls showed Stevens performing better, and El-Sayed's one-point primary win over Stevens, despite being heavily favored, signals weak consolidation of the Democratic base. This candidate-specific drag tempers the national tailwind. — DeepSeek (79% on 2,145 graded calls)
Notably, not one seat produced a lopsided number in either direction. The widest gap on the board after the revision round was Fable at 57% against GLM and DeepSeek at 45%, which is a twelve-point spread on a question where the market sits 12.6 points above their blend and between 5.5 and 17.5 points above any individual seat.
Where The Panel Changed Its Mind
Three seats moved after reading the others, and the movements ran in both directions, which is the useful kind.
Gemini made the largest revision on this board, cutting El-Sayed from 58% to 48%. It had started by treating the environment as close to decisive and came back weighing the counterfactual evidence it had underrated:
While the D+6.6 generic ballot and adverse presidential approval provide a strong structural tailwind, the counterfactual polling showing Stevens consistently outperforming El-Sayed reveals a measurable candidate-specific drag that offsets much of this environmental advantage. — Gemini (85% on 2,062 graded calls)
GLM moved the other way, from 40% to 45%, persuaded by an argument about who is still undecided:
The structural case is stronger than I credited: Slotkin beat this same Rogers in a Trump-carrying 2024 environment, so a D+6.6 generic ballot with 11.6% undecideds, which historically break toward the out-party, should lift even a candidate-specifically weakened Democrat closer to even. — GLM (81% on 2,138 graded calls)
DeepSeek held at 45% and said exactly why, which is worth as much as a revision:
I held my estimate because no peer surfaced new evidence or reasoning I had not already weighed; the higher estimates rely on the same national environment and base-rate argument, which I already discounted given the candidate-specific drag and lack of post-primary data. — DeepSeek
The round did what convergence rounds are supposed to do. The spread across the eight seats narrowed from 18 points to 12, and the blend barely moved, from 50.3% to 49.9%. The models talked themselves closer together without talking themselves into the market's number.
Why The Market Still Has El-Sayed At 62 Cents
The market is not being careless, and the panel did not treat it that way. Three arguments hold up its price, and every seat weighed all three.
The first is that the summer polls have not met the general election yet. Almost every head-to-head in the record was taken while Democrats were still fighting each other, which is the worst possible moment to measure how a Democratic nominee performs against a Republican. Party voters consolidate after primaries, sometimes by several points, and both of El-Sayed's rivals endorsed him inside twelve hours.
The second is the environment, which is the strongest single number on the board. A D+6.6 generic ballot in a state that elected a Democratic senator during a Trump presidential win is a serious tailwind, and midterm polling has been close to unbiased in the Trump era, running −0.3 in 2018 and +0.9 in 2022. There is no history here of polls systematically understating Democrats in midterms, but there is plenty of history of the president's party losing ground.
The third is that undecideds have somewhere to go. Roughly 11.6% of voters are unallocated in the aggregate, which is a large pool this late, and out-party candidates have historically collected more of it in a referendum midterm.
This is a real disagreement about which signal dominates: the national environment or the candidate-specific evidence in front of us. Both sides of it are supported by fetched data, and the raters who do this professionally split the same way. It is also not the only Midwest board where the panel sits below a confident Democratic price: on Minnesota's open Senate seat, the models likewise shave several points off a near-certain DFL price.
Dark Horses The Panel Won't Dismiss
The board looks like two names now, but the fetched record leaves scenarios worth naming.
- A Consolidation Bounce That Makes The Market Right And The Panel Wrong. If the first post-primary poll shows El-Sayed running even or ahead, that supports the market's consolidation case and should pull the panel back toward 62 cents. Every seat on the board named this as the live risk to its own estimate. Fable and Opus both said the first post-primary head-to-head is the single most informative thing that will happen to this race.
- A Drag That Turns Out To Be Structural. The opposite scenario is that El-Sayed's deficit was never about the primary being contested. If mid-August polling shows the same two-to-four-point gap that existed in June, the models' number looks conservative rather than skeptical, and a market still pricing 62 cents would be the outlier.
- The Two Contracts That Are Already Decided. Kalshi's own rules say a listed candidate who fails to secure the nomination resolves No immediately, which is now true of both Stevens and McMorrow. As of this fetch the exchange still shows both as active, bid at zero with a one-cent ask, and Stevens's line still carries about 97,000 in open interest, so the settlement has not been processed yet. They are a reminder that a listed contract is not the same thing as a live outcome.
What Would Change The Panel's Mind
Four things, all of them checkable, in the order the models ranked them.
- The first post-primary head-to-head poll, likely mid-to-late August. Named by seven of the eight seats as the most informative upcoming indicator. El-Sayed closing to even or better pushes the blend toward the market; the same gap persisting pushes it further away.
- A rating agency updating the race. Every current rating predates the primary. Cook or Sabato moving off Toss-up in either direction would be the first professional judgment on the matchup that exists.
- A move in the generic ballot. The panel's floor under El-Sayed is the D+6.6 environment. Gemini specifically flagged generic-ballot movement as the number that would most change its estimate, in either direction.
- Third-quarter fundraising, reported in October. Through June 30, El-Sayed had raised $14.5 million and had $2.55 million on hand after a bruising primary; Rogers had raised $10.9 million and sat on $4.47 million, per FEC filings. Whether El-Sayed refills that account after a costly win is a real test of the consolidation thesis.
Settlement Timeline
| Date | What happens |
|---|---|
| August 4, 2026 | Democratic primary. Settled: El-Sayed's nominee contract resolved YES on Kalshi |
| Mid-To-Late August 2026 | First post-primary general-election polling expected. The page re-scores if the number moves |
| October 2026 | Third-quarter FEC reports; final rating-agency updates |
| November 3, 2026 | Election day. KXMISENATE-26 expires and settles on the winner |
This page is re-scored whenever the story moves, and every price carries the timestamp of the run that fetched it. The current stamp is August 5, 2026, 10:45 a.m. ET.
The Traps
Transparency matters more than a clean story, so here is where this board could mislead you.
- A Coin Flip Is A Finding, Not A Dodge. Eight seats landing at a 49.9% blend is not the panel refusing to commit. Individual numbers ranged from 45% to 57%, each with a stated mechanism, and they converged on "too close to call" from opposite directions. The interesting claim is not that Michigan is close; it is that the market prices a clear favorite in a race the models score even. For the mirror image, see Ohio's Husted-vs-Brown board, where the models land on the same coin-flip read and the market essentially agrees with them.
- Nobody Has Polled This Matchup Since It Became Real. Every head-to-head on this page was taken before Michigan Democrats voted. That cuts against the panel's confidence as much as the market's, and it is the reason both numbers should be treated as provisional for another few weeks.
- A 12-Point Gap Is A Disagreement, Not A Proven Edge. Two methods reading the same race differently is a signal to look harder, not proof that either is wrong. Prediction markets on state Senate races carry real spreads, and the quote fields on this book were null through the night, so the mid prices here are drawn from the exchange's hourly candle record rather than a live two-sided quote.
- The Primary Count Is Unofficial. El-Sayed's win rests on the reported count and a concession, and Kalshi has settled its nominee market accordingly, but the state canvass has not certified the result.
How To Read These Prices
A 62.5-cent YES price implies roughly a 62.5% chance El-Sayed wins the seat. To get a fair read you normalize the field so the prices sum to 100. On this board the four contracts add to about 103 cents, so El-Sayed's fair number is closer to 60.7% and Rogers's to 38.3%. The favorite is almost always a shade softer than the raw price suggests, which is the whole idea behind converting Kalshi's cents into real odds before comparing them to anything.
That habit of normalizing a price before comparing it to an independent estimate is the whole method behind this series, and it is the one our Texas Senate model verdict applies to a very different field.
OddsShopper's odds screen and free expert picks cover the sports slate rather than election futures, and the Michigan race itself lives on Kalshi, but the discipline is the same, and every number on this page lands in our publicly graded Model Verdict ledger.
A few honest points before reading too much into any number above:
- Eligibility. Kalshi is a CFTC-regulated event-contract exchange, not a sportsbook. You must be 18 or older and located in an eligible state, and availability varies by state and can change.
- Liquidity, Spread, And Fees. This book has real open interest but quoted prices that go dark for hours at a time. Your fill depends on the spread at the moment you act, and Kalshi charges trading fees on some markets that eat into any edge.
- A Moving Story. Three months remain, no post-primary poll exists yet, and the estimates here reflect only what was known on the morning of August 5.
- Not Advice. Everything here is analysis for information, not a recommendation to trade or to vote. These are model estimates, not predictions of fact and not financial or voting advice.
Hottest Prediction Markets Right Now
- 2028 Democratic presidential nominee · $169M traded
- 2028 U.S. Presidential Election winner? · $57M traded
- 2027 Pro Football Champion · $54M traded
- Pro Baseball Champion · $45M traded
- More tech layoffs in 2026 than in 2025? · $31M traded
- Bitcoin price at the end of 2026 · $30M 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
Michigan Democrats picked the nominee who polled worse against Mike Rogers in every survey taken before the vote, and they picked him by a single point after the polls said fifteen. The exchange absorbed all of that in one night, dropped El-Sayed twelve cents at the tightest point of the count, and then put him back within three cents of where he started. Our eight-model panel, reading the same fetched record without seeing a price, cannot get him past even: 49.9% against a 62.5-cent market, with the spread between the models narrowing after they argued and the blend not moving at all. Cook and Sabato called this a toss-up months ago, before anyone knew who the Democrat would be. Tuesday gave everyone new information about the Democratic nominee, and it did not make the race any less close. If this method is your thing, our North Carolina Senate board and our 2026 Senate election odds chamber breakdown are the natural next reads.
