Florida's primary night settled the question everyone was trading and exposed the ones nobody was. Byron Donalds won the Republican nomination for governor, David Jolly won the Democratic one, and the Associated Press called both races before most of the vote was in. The contracts that paid on those two names had been trading at or above 98 cents since the first week of August. The contracts that did not pay were the interesting ones, and they were wrong in a pattern worth naming before anyone stakes an opinion on November. We graded every closing price against the count, then handed the general election to an eight-model AI panel, price-blind, with the state's fetched record in front of it.
The Quick Answer
Eight price-blind AI models make the Republican side a 78.5% favorite to hold the Florida governorship against a market price of 80.5 cents, with the Democratic side at 20.9% against 20 cents. That is agreement, not an edge, and the agreement is the finding: after a revision round in which every seat read the others' anonymized reasoning, the panel's spread narrowed from 18.5 points to ten and landed within two points of the crowd. The primary-night grading underneath it is where the markets actually got caught, and the full board, the seat-by-seat numbers, and the four things that would reprice this race are below.
Kalshi's market prices a party, not a person, and it is asking a narrower question than the headlines suggest. At 80.5 cents, the exchange is quoting roughly a four-to-one shot, the equivalent of a sportsbook moneyline of about −413 on the Republican side (risk $413 to win $100) and about +400 on the Democratic one (risk $100 to win $400). That is a clear favorite with a real losing branch, which is a different animal from the near-lock the primary market had become.
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.
New to event markets? How to bet sports on Kalshi explains how these contracts settle and what a price in cents actually means. Prices on this page refresh on our standard market-pulse cycle, with the as-of stamp below carrying the evidence.
The Board
| Outcome | Nominee | Kalshi | AI Blend (after revision) |
|---|---|---|---|
| Republican | Byron Donalds | 80.5¢ | 78.5% |
| Democratic | David Jolly | 20.0¢ | 20.9% |
| Neither party | independent or minor-party win | not listed | 1.0% |
These are model estimates, not predictions of fact and not financial advice.
More live boards from the same panel: The 2026 Senate hub, where Florida's Senate special sits at 91¢ Republican · Georgia governor · Bottoms 61¢ · Michigan governor · Benson 85¢. Prices fetched August 18, 2026.
Hottest Prediction Markets Right Now
- 2028 Democratic presidential nominee · $185M traded
- 2027 Pro Football Champion · $66M traded
- 2028 U.S. Presidential Election winner? · $60M traded
- 2028 Republican presidential nominee · $58M traded
- Pro Baseball Champion · $56M traded
- More tech layoffs in 2026 than in 2025? · $31M 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 Market
Two contracts are listed. GOVPARTYFL-26-R resolves YES if a Republican is inaugurated as governor of Florida pursuant to the 2026 election; GOVPARTYFL-26-D resolves YES if a Democrat is. Settlement keys on inauguration, with an expected expiration of January 8, 2027, though the rules allow an accelerated determination once a consensus of media organizations projects the winner. About 236,600 contracts have matched across the pair. Polymarket runs the same question with far less money behind it and lands in the same place: 82 cents Republican, 18.5 cents Democratic, on about $36,000 of volume.
Two details in that language matter more than they look. First, the contracts settle on PARTY, not on a person, so nothing on this board is a bet on Donalds or Jolly individually. Second, because only two party contracts exist, a win by one of the qualified independents on Florida's November ballot would resolve both of them NO. The panel priced that path at about 1%, and the market does not quote it at all.
Two timestamps: the panel's verdicts were generated the evening of August 18, 2026, price-blind and frozen since; prices were fetched from the exchanges the same evening, about 9:44 p.m. ET. Kalshi is a CFTC-regulated event-contract exchange, not a sportsbook; 18+, availability varies by state as of August 2026.
How The Primary Markets Actually Graded
The nominee contracts were never in doubt and never claimed to be. On Kalshi, Donalds closed at 99.9 cents and Jolly at 99.7; on Polymarket, both closed at 99.5. Kalshi has since marked both nomination contracts finalized, resolving Donalds and Jolly YES and every rival NO; Polymarket's versions are still open and trading at a penny either side of certainty. Grading a 99-cent favorite is not a test of anything.
The tests were the side markets built on top of the main one, the contracts asking by how much and in what order rather than simply who, and they failed in a consistent direction. Polymarket's margin-of-victory board closed with "Donalds wins by 40 points or more" as its favorite at 47.5 cents and "35 to 40 points" second at 24.5 cents. The band the unofficial count landed in, "20 to 25 points," was trading at three cents at the close. In unofficial returns, Donalds finished at 48% and Jay Collins at 25%, a provisional margin of roughly 23 points. A trader who bought the correct band at the close paid three cents for a dollar.
The same error shows up on the vote-share board. "James Fishback under 10%" was the market's favorite at 49.5 cents when polls closed. In unofficial returns Fishback took about 175,000 votes, roughly 10.5% of the Republican ballots counted so far, which puts him in the 10-to-15 band that was trading at 28.5 cents. Worse for the market, it had been drifting the wrong way for two solid weeks: the "under 10%" line sat at 18.5 cents on August 6 and climbed steadily to 52.5 cents by election morning. The market moved away from the right answer as the vote got closer, not toward it.
The second-place board got the answer right and sweated it anyway. Collins was the runner-up favorite all month, reached 82.5 cents twenty-four hours before the close, and then slid to 70.5 cents at the close as Fishback's second-place price nearly doubled from 17.5 to 30.5 on election day itself. In unofficial returns Collins finished second, about 246,000 votes clear of Fishback. The board that was right was least confident at the exact moment it should have been most confident.
There is a knowable cause underneath all three misses, and it is not the market's alone. The public polling aggregate for this primary, per 270toWin's Florida governor page covering fields from July 13 to August 6, had Donalds at 47.5%, Collins at 13.5%, Fishback at 9.0% and Renner at 4.5%. Donalds finished at 48%, half a point from the aggregate. Collins finished at 25%, nearly twelve points above it. The polls measured the frontrunner almost exactly and badly missed where the enormous undecided bloc would go, and the market inherited that error wholesale. This is the general lesson of the night: prediction markets are strongest on the question the polls are strongest on, and they compound the polls' weakness everywhere else.
For the record, our own calls graded honestly, and one of them graded badly. Our Florida governor primary odds page named Collins the 77-cent second-place favorite, and Collins finished second.
Our primary model verdict blended Donalds to about 82% against a 99-cent market price. He won. The crowd was better calibrated than our panel on that one, and we take the loss on the scoreboard rather than quietly dropping it.
Our Fishback page reported the even-money line on his vote share sitting just under 10% on the evening of the vote. He landed just above it. Every one of those calls was in the ledger before the polls closed.
Byron Donalds: The Trump-Endorsed Favorite Who Took 48%
Donalds has represented Florida's 19th district in Congress since 2021 and has carried President Trump's endorsement since February 2025, per the race's public record. He raised $13.79 million and spent $5.38 million through August 13, per Florida Division of Elections filings, roughly seven and a half times what Collins raised. He is also the first Republican in eight years to run for this office without an incumbent's shield, and he has never run statewide.
The 48% is the number the panel kept returning to. Winning with under half the vote is normal in an eleven-candidate field, but it means Donalds starts the general needing to consolidate the more than 830,000 Republicans who, on Tuesday, voted for someone else. Claude Fable moved its number down specifically on that point, and the reasoning is quoted in full below.
Donalds also arrives with an attack to answer, which is worth reporting plainly rather than leaving in the subtext of a probability. In the closing weeks of the primary, rival candidate James Fishback posted on X that a federal investigation into Rep. Cory Mills had expanded to include Donalds. On August 7, 2026, a Department of Justice spokesperson told Newsweek that "a law enforcement official confirms this is absolutely false"; Newsweek reported the same day that the documents Fishback posted had not been independently authenticated, and quoted Donalds spokesperson Gates McGavick calling the claim a "baseless and defamatory sideshow." Donalds' contract barely registered any of it. Between August 5 and the close, his nomination price never traded below 96.5 cents on either exchange, and the count agreed with the price.
David Jolly: The Republican Who Switched Sides
Jolly represented Florida's 13th district as a Republican from 2014 to 2017, lost the seat to Charlie Crist, published a 2018 NBC opinion piece arguing that "you can't be never-Trump and be a Republican" (Ballotpedia's record of his career), and registered as a Democrat in April 2025. In unofficial returns he took 61% of a six-candidate Democratic primary, the cleanest mandate either nominee carries into November.
His central pitch is that a former Republican can pull disaffected Republicans across, and the polling record gives that pitch partial support: both head-to-heads that show him ahead are Democratic-sponsored reads, and the university polls (North Florida, Stetson, Emerson) have had this race inside seven points all year. The counterweight is his own file. CBS News Miami reported on August 3, 2025 that Jolly's sharpest criticism was coming from fellow Democrats rather than Republicans, chiefly on abortion: as a Republican congressman he said he believed life begins at conception, and he now says his position has changed and he supports abortion rights. The panel weighed that as a base-turnout question rather than a persuasion one.
What The Fetched Record Says
Seventeen public Donalds-versus-Jolly head-to-heads exist, going back to August 2025. The complete list is on the race's polling page, and the shape is stable: Donalds leads clearly in eleven, Jolly leads clearly in two, two are dead ties, and two split depending on which ballot version or voter screen you read (a May Change Research field had Jolly by four among likely voters and Donalds by two among registered voters). The Donalds leads mostly cluster between four and eight points, with one eleven-point outlier last October. The university polls sit in a narrow band. The University of North Florida had it Donalds 46, Jolly 41 in mid-July. Stetson University had Donalds 47, Jolly 40 in the spring. Emerson had Donalds 44, Jolly 39 in March.
The freshest poll is also the one that cuts the other way. Hart Research, sponsored by the Democratic Governors Association, had Jolly 46, Donalds 45 on August 10 to 13, inside a four-point margin of error. It is the only August number in the file and the only 2026 Jolly lead outside Change Research, a Democratic firm. Every seat on the panel discounted it for sponsorship and none of them dismissed it, which is roughly the right handling of a partisan poll that agrees with a plausible story.
The structural block underneath the polling is heavy and points the other way. No Democrat has been elected governor of Florida since 1994. No Democrat has carried the state for president since 2012. Republicans hold every statewide office, both U.S. Senate seats and supermajorities in both legislative chambers. DeSantis won re-election by 19.4 points in 2022, and Trump carried the state by 13 in 2024, up from 3.4 in 2020.
And then there is the one Democratic near-miss, which is the most instructive fact on the card. The last open-seat Florida governor race, in 2018, was decided by four tenths of a point. That result had three causes worth checking against today rather than citing as vague caution: a national blue wave in a first-term Republican president's midterm, an open seat with no incumbent on the ballot, and an untested statewide Republican nominee. The first two conditions are present again. The generic congressional ballot averages Democrats +6.6 across six aggregators as of August 17, presidential approval sits at 37.6% against 59.3% disapproval, and the president's party has lost House seats in eighteen of twenty postwar midterms. The third is arguable, since Donalds has won House races but never a statewide one. What has changed since 2018 is the baseline itself: Florida's voter registration advantage flipped Republican, and the presidential margin moved nearly ten points right.
The primary turnout numbers are the last piece, and they are noisier than they look. On the votes counted so far, about 1.64 million Republicans voted for governor against about 1.23 million Democrats, a Republican electorate roughly a third larger. These are closed primaries, so that measures each party's registered and motivated base rather than the November electorate, and the two contests had different levels of drama. Two seats on the panel raised exactly that objection and one revised its number because of it.
Every Seat's Number
Both rounds shown, Republican-side probability. The headline blend is the median of the revised column. On a phone this table scrolls sideways.
| Seat | R1: Republican | R2: Republican | R2: Democratic | R2: Neither |
|---|---|---|---|---|
| Claude Opus | 88.5 | 85.0 | 14.3 | 0.7 |
| Claude Fable | 83.0 | 82.0 | 17.0 | 1.0 |
| Gemini 3.1 Pro | 82.0 | 80.0 | 19.5 | 0.5 |
| Claude Sonnet | 79.0 | 79.0 | 20.0 | 1.0 |
| DeepSeek V4 | 78.0 | 78.0 | 21.8 | 0.2 |
| ChatGPT (GPT-5.5) | 75.0 | 77.0 | 22.0 | 1.0 |
| GLM 5.2 | 70.0 | 75.0 | 24.0 | 1.0 |
| Kimi K3 | 70.0 | 75.0 | 24.0 | 1.0 |
Every seat on this panel is graded against real market settlements — records to date: GPT 84% on 5,500 graded calls · Claude Fable 84% on 508 graded calls · Claude Opus 85% on 578 graded calls · Claude Sonnet 84% on 556 graded calls · Gemini 86% on 4,292 graded calls · Kimi 83% on 2,345 graded calls · GLM 81% on 2,367 graded calls · DeepSeek 80% on 2,392 graded calls. Recomputed daily; the full scoreboard is public.
Model estimates generated August 18, 2026, price-blind from live fetched data. These are model estimates, not predictions of fact and not financial or trading advice. Models are frequently wrong; the market price reflects real traders' money. Kalshi is a CFTC-regulated event-contract exchange; 18+, availability varies by state.
Every number in this piece landed in our backtest ledger the evening it was generated and gets graded against the certified result on the public model verdict scoreboard.
Where The Panel Changed Its Mind
Round one ran from 70 to 88.5, an 18.5-point spread. The revision round closed it to ten, and six of eight seats moved. The two ends walked toward each other for stated reasons, and the middle held for a stated reason.
The top of the range came down the hardest, and the argument it accepted was about double-counting:
"I accepted the point that I was double-counting structure. My 88.5 implied roughly a ten-point win with tiny variance, but the actual polling median across the head-to-heads is four to six points, and Trump-era midterm polls were nearly unbiased. A five-point median lead with eleven weeks left, an open seat, a Democrats +6.6 generic ballot and 37.6% approval is 'clear favorite,' not 'near-lock.'" (Claude Opus, 88.5 to 85.0)
The bottom of the range came up on the mirror-image argument:
"I moved from 70 to 75, accepting two specific peer mechanisms. First, Trump-era midterm polls were nearly unbiased, and sixteen of seventeen head-to-heads cluster at Republican +4 to +7 across diverse sponsors; the honest point estimate is about +5, and a five-point August deficit with ten weeks left is historically a 20-to-25% comeback proposition, not 28%. Second, Florida delivered a 19-point Republican governor win in 2022, a nationally neutral-to-Democratic year, which shows the state's rightward drift resists national waves." (Kimi K3, 70.0 to 75.0)
[Editor's note: the seat's "sixteen of seventeen" count is its own reading, not ours. On the fetched record the tighter description is eleven clear Donalds leads out of seventeen, most of them between four and eight points. The direction of the seat's argument survives the correction; the count does not. Model rationales are published as written.]
The single most specific revision came from the fact nobody else had weighed:
"Shaved one point for the consolidation observation: Donalds won his primary with only 48%, an untested-statewide nominee needing to unify Collins and Fishback voters. A Trump endorsement usually heals that quickly, but 'usually' isn't 'certainly' for a first-time statewide candidate." (Claude Fable, 83.0 to 82.0)
Two seats did not move, and one of them explained why in a way worth quoting:
"No peer introduced new evidence, only re-weighted the same polling-versus-structure tension. The lower numbers lean harder on the 2018 open-seat parallel, but Florida's baseline has shifted materially rightward since then, which is the specific reason 2018 shouldn't fully repeat." (Claude Sonnet, held at 79.0)
What Would Change The Panel's Mind
- The first post-primary public poll. Every head-to-head in the record was fielded before Tuesday's vote. Not one was taken with the field final and both nominees known. A nonpartisan read showing this inside three points would push the panel's number down several points; a read at Republican +8 or better would push it up.
- Whether Collins and Fishback endorse, and how fast. More than 830,000 Republicans voted for one of his rivals. Public consolidation inside two or three weeks reads as a normal primary healing and holds the number; a public holdout past Labor Day is the specific event that makes 48% look like a ceiling rather than a floor.
- Third-quarter fundraising, reported in October 2026. Donalds ended the primary having raised about seven and a half times what his nearest rival did. If Jolly's first general-election quarter closes that gap materially, the panel's Democratic number moves; if the gap widens, it does not.
- A three-point move in the national numbers. The panel's Democratic case leans on a Democrats +6.6 generic ballot and 37.6% presidential approval. Both are measured weekly by public aggregators, and a three-point shift in either direction is the cheapest thing on this list to check.
Settlement Timeline
- November 3, 2026 — election day. The contracts can settle early on an accelerated determination once a consensus of media organizations projects a winner.
- Late August to mid-September 2026 — county canvassing boards certify the primary, and the first post-primary head-to-head polls should land. Both are checkable dates.
- October 2026 — third-quarter campaign finance reports, and debate season if debates are scheduled.
- January 8, 2027 — the contract's expected expiration, keyed to inauguration.
- This page re-scores when the story moves. Prices above are as of August 18, 2026, about 9:44 p.m. ET.
How To Read These Prices
- Eligibility. Kalshi is a CFTC-regulated event-contract exchange, not a sportsbook. 18+, availability varies by state as of August 2026. Election event contracts are legal to trade there.
- Liquidity. About 236,600 contracts have matched across the two listed markets, split almost evenly between them. The gap between the best buy price and the best sell price runs three cents on both sides, wide enough that the price you can actually transact at is not the price on the screen.
- The sum rule. The two midpoints add to 100.5 rather than 100, and that extra half-cent is the exchange's built-in margin. The panel's three medians add to 100.4 for a different reason: each outcome's median is taken independently, so they are not forced to sum to exactly 100.
- The count is provisional. Every vote figure on this page comes from election-night returns at roughly 94% of precincts, cross-checked across two outlets. Florida had not posted certified results at the time of writing.
- Not advice. These are model estimates, not predictions of fact and not financial advice.
The Bottom Line
The market and the panel agree that a Republican holds Florida, and they agree by almost exactly the same amount, which is not a trade. The value in primary night is somewhere else. The exchanges nailed the two questions the polls had already answered and misfired badly on every question that required judgment about the undecided pool, pricing a 40-point Donalds blowout at 47.5 cents and the band the unofficial count actually landed in at three, and drifting away from the right answer on Fishback's vote share for two straight weeks. A market that is only as good as its inputs is still useful, but it is not oracular, and the difference matters most in exactly the places people find most tempting. November's question has no polls yet at all. When the first one arrives, this page re-scores, and the same panel that reads Georgia's open governor race as a near coin flip, with its favorite at 61 cents against Florida's 80.5, will grade in public either way.
