Florida Republican Governor 2026: The Primary AI Model Verdict Vs. The Kalshi Market

Florida's Republican primary for governor is Tuesday, August 18, 2026, an open-seat race to succeed term-limited Gov. Ron DeSantis. Kalshi has turned it into a market, one yes/no contract per candidate, resolving YES only for whoever wins the nomination. We asked six AI models to score the entire field without letting any of them see a single price. They came back with a clear front-runner, U.S. Rep. Byron Donalds, but with noticeably more doubt than the 97-cent market is showing, and with a far bigger opening for Lt. Gov. Jay Collins than his sub-penny price implies.

New to Kalshi event markets? Start here → how Kalshi's yes/no contracts work.

The Quick Answer

Kalshi's market makes Byron Donalds the heavy favorite near 97 cents, treating the nomination as all but decided. Our price-blind, six-model panel leans the same way but far less hard: it blends Donalds to about 82%, and it gives Jay Collins roughly 14%, against a market price of under one cent. In other words, the models see the same front-runner as the crowd, but they price in a rising second-place challenger, a large undecided bloc, and Florida's plurality-wins rule as real sources of uncertainty that a near-97-cent price does not.

• Market Front-Runner: Byron Donalds, ~97¢
• Model Front-Runner: Byron Donalds, 81.6% blend (still first, but well short of the market's near-certainty)
• Where The Models Diverge Most From The Price: Jay Collins, a ~14% blend against a market price under 1¢ (they also trim the front-runner, from ~97¢ to an 82% blend)
• The Unallocated Bloc: roughly 28 to 31 percent of two July polls sat with undecided voters, still in play with the primary two weeks out
• The Panel's Logic In One Line: a lead in every July poll plus Trump's endorsement plus a plurality rule is hard to beat, but one rising challenger and a big undecided bloc are not the near-certainty a 97¢ price suggests.

The case for Donalds, where the models part from the market on Collins, the full candidate board, the dark horses, and a plain-English guide to reading these prices are all below.

The Market At A Glance

Venue	Kalshi, a CFTC-regulated event-contract exchange (18+; availability varies by state, as of August 2026)
The Event	"Florida Republican Governor nominee?" (Kalshi event KXGOVFLNOMR-26), one yes/no contract per candidate
The Contract	Each market resolves YES only for the candidate who wins the Republican nomination for governor at the August 18, 2026 primary, per Kalshi's settlement rules
Window	The primary is August 18, 2026; the board's listed expected expiration is November 3, 2026, and contracts settle after the nomination is decided
Prices Below As Of	August 3, 2026 (latest snapshot; prices on this page refresh during the news wave into the primary)

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. Every number in this piece gets graded in public once the market settles, in the full graded Model Verdict scoreboard.

What Actually Decides This Primary

Before any name, the shape of the race matters, and it is what kept the models from matching the market's near-certainty. Three fetched facts define the board:

• It Is An Open Seat With A Trump-Endorsed Front-Runner. Two-term Gov. Ron DeSantis is term-limited and cannot run again. U.S. Rep. Byron Donalds entered the race with President Trump's endorsement, first given in February 2025 and repeated since, and has added backing from Sen. Rick Scott and House Speaker Mike Johnson. He has led every public poll and holds a fundraising edge.
• The Rules Reward A Plurality, Not A Majority. Florida abolished primary runoffs in 2005; the nominee is whoever wins the most votes, even under 50 percent. In a fragmented field, that structurally protects a consistent front-runner, because rivals splitting the anti-front-runner vote cannot force a second round.
• A Second Name Is Rising, And A Big Bloc Is Still Undecided. Lt. Gov. Jay Collins, appointed by DeSantis in 2025 and a former Green Beret, entered the race in January 2026 and climbed in the July polls after a multimillion-dollar ad blitz. Across July surveys the undecided share ran about 28 to 31 percent, and where it lands is unknown.

That framework is the whole story of this board. The market is pricing Donalds's lead as close to settled. The models are pricing the same lead, minus a discount for a rising challenger and a big undecided bloc in a race that is still two weeks from a vote.

Here is the short version first, then every model's raw number.

The Board

Candidate	Status	Market	Model blend
Byron Donalds	Active (Trump-endorsed, poll leader)	~97¢	81.6%
Jay Collins	Active (Lt. Gov., rising in polls)	<1¢	13.6%
James Fishback	Active (investor/activist)	~3¢	2.7%
Paul Renner	Active (ex-House Speaker)	<1¢	1.0%
Casey DeSantis	No confirmed 2026 campaign	<1¢	0.2%
Matt Gaetz	No confirmed 2026 campaign	<1¢	0.0%
Wilton Simpson	No confirmed 2026 campaign	<1¢	0.0%
Jimmy Patronis	No confirmed 2026 campaign	<1¢	0.0%
Charles Burkett	No confirmed 2026 campaign	<1¢	0.0%

Every Seat's Number

Candidate	Market	Fable	Opus	Sonnet	GLM	Kimi	DeepSeek	Blend	Gap (blend% − mkt¢)
Byron Donalds	~97¢	86.7	84.0	78.8	75.0	80.0	85.0	81.6%	-15
Jay Collins	<1¢	10.2	12.0	14.1	18.0	15.0	12.0	13.6%	+14
James Fishback	~3¢	2.0	2.0	4.0	3.0	3.0	2.0	2.7%	0
Paul Renner	<1¢	0.0	1.0	2.0	2.0	1.0	0.0	1.0%	+1
Casey DeSantis	<1¢	0.0	0.0	0.0	1.0	0.0	0.0	0.2%	0
Matt Gaetz	<1¢	0.0	0.0	0.0	0.0	0.0	0.0	0.0%	0
Wilton Simpson	<1¢	0.0	0.0	0.0	0.0	0.0	0.0	0.0%	0
Jimmy Patronis	<1¢	0.0	0.0	0.0	0.0	0.0	0.0	0.0%	0
Charles Burkett	<1¢	0.0	0.0	0.0	0.0	0.0	0.0	0.0%	0

These are model estimates, not predictions of fact and not financial or voting advice. The Blend column is the simple mean of the six model numbers shown in its row. The last column is a quick directional gap, the blend percentage minus the raw market price in cents; it compares a model probability to an un-adjusted price, so read it as direction and size, not a precise edge. Market figures are raw last-trade YES prices; unusually for a prediction market, this board's raw prices already sum to about 100 cents, so there is little vig to remove. Below the top three, most models did not name an explicit number, so the small values are each model's leftover "field" mass; treat anything under about 1% as noise, not a distinct read, and note that several of those names have no confirmed 2026 campaign and price near zero.

A note on the panel. Our standard Model Verdict lineup runs eight seats. On this run two were unavailable: the ChatGPT and Gemini seats were not installed on the machine that produced these verdicts. This board therefore runs on six models: Claude Fable, Claude Opus, Claude Sonnet, GLM, Kimi, and DeepSeek. All six scored the field on August 3, 2026, without seeing any prices, working from a fetched, dated summary of the candidate field, the candidacy timeline, and the July 2026 polls. They then read one another's reasoning anonymized and revised, and the numbers here are the post-revision blend. Every model's number also lands in our back-tested Model Verdict ledger, where past price-blind calls are graded against settled results so the method is measured, not just asserted.

The board really comes down to three things: the front-runner the models still back, the second name where they part hardest from the market, and the undecided bloc that makes this less settled than the price says.

The one line that explains the whole board: the market treats Donalds's lead as nearly decisive; the models treat it as a strong lead in a race with a rising challenger and a third of the vote still undecided. Both make him the favorite, but only one of them calls it 97 cents.

Byron Donalds: The Front-Runner, Market And Models Agree

Every model made Donalds the favorite, and no model put him below 75%. His case is the polling and the structure of the race. Kimi laid out the arithmetic most plainly: "his minimum within-poll lead over the runner-up is ~16 points," and "Florida's no-runoff plurality rule means he needs only ~30-35% of the vote in a fragmented field, not 50%." Kimi added the cleanest summary of why the lead is durable: "Donalds's worst polled number (37.5%) still exceeds Collins's best (30%), so a Collins win requires both large late movement AND systematic polling error in the same direction."

Fable made the same structural point about the calendar: even "the most Collins-favorable snapshot (Wolfson, Jul 30: 46-30) still shows Donalds up 16 points late in the race, and a challenger closing a mid-teens gap in two weeks against a Trump-endorsed front-runner" is a tall order. GLM, which landed as the most cautious seat on Donalds at 75%, still granted the late-decider dynamic cuts in his favor, noting that "late deciders in GOP primaries tend to break toward the Trump-endorsed frontrunner."

Where the panel held back, in its own words: the models did not treat the lead as safe. Sonnet, at about 79%, kept Collins as "the real risk to this pick," citing "the Lt. Gov. appointment, multimillion-dollar ad blitz, and the Wolfson memo" showing "the gap tightening to 16pts with a huge ~28-31% undecided bloc still in play." That uncertainty is exactly the gap between an 82% blend and a 97-cent price.

Jay Collins: Where The Models Part From The Market

This is the sharpest split on the board. The market prices Collins under one cent; the panel blends him to about 14%, the sharpest gap on the board where the models price a candidate above the market. The models are not predicting he wins, they are saying the crowd's sub-penny price ignores a real, if secondary, path.

The reasoning was consistent across seats. DeepSeek called Collins "the only rising challenger," and Opus called him "the only candidate with a plausible path, a sitting Lt. Gov. with money and momentum," which is why "nearly all of the residual" probability mass landed on him. Opus's own arc is the clearest revision on the board: after reading the other seats, it wrote that it "hadn't fully priced the Wolfson 30% reading plus the ~30% undecided bloc" as real variance under plurality rules, and lowered Donalds from 89% to 84% while roughly doubling Collins. GLM moved the same direction on Collins's late movement, holding "below the 80s because the Wolfson poll's Collins-at-30% reading" reflects real late movement "that the higher-probability forecasters may be underweighting." The models still keep Collins a clear second because, in their repeated caveat, every single fetched poll still had Donalds first, several by wide margins.

Dark Horses The Panel Won't Dismiss

Below Collins, the field thins fast, but a few names drew a real number or a named reason from at least one seat:

• James Fishback (Blend 2.7%, Market ~3¢). The investor and activist is the one lower-tier name the market and the models actually agree on. He polled in the high single digits to low double digits across July surveys, and several seats gave him a small but non-zero number. His model blend and his market price line up almost exactly, so there is no hidden edge here, just an honest longshot.
• Paul Renner (Blend 1.0%, Market <1¢). The former Florida House Speaker is a credentialed name running an active campaign, and Opus, Sonnet, GLM, and Kimi each left him a point or two of probability. He polled in the low single digits, so the models treat him as a live-but-distant longshot rather than a real threat.
• Casey DeSantis (Blend 0.2%, Market <1¢). Only GLM gave her even one point, on the theory that a late entry by the sitting first lady would reset the race. The rest of the panel treated the absence of a confirmed 2026 campaign as decisive and priced her at essentially zero.

The Traps: Rows That Look Like Signal But Are Not

Transparency matters more than a clean story, so here is where the table would mislead you:

• The No-Campaign Names Cannot Be Assumed Live. Casey DeSantis, Matt Gaetz, Wilton Simpson, Jimmy Patronis, and Charles Burkett all have listed contracts, and a couple carry tiny blend numbers, but none is a polled 2026 candidate in the fetched record. Those small values are leftover "field" mass, not real paths, and their prices sit at the exchange's one-cent floor.
• A Sub-Penny Price Is Not The Same As Zero Probability. Collins trades under a cent, but that is a thin-market price, not a measured 0% chance. The whole point of the panel's 14% blend is that a second-place, well-funded sitting lieutenant governor is not a rounding error, even if the order book treats him like one.
• The Bottom Of The Table Is Noise. With a six-model panel and most names unscored, do not read a real signal into a 0.2% row.

How To Read This Board

The board sorts into three honest reads, and which one fits depends on how much weight you put on the plurality math versus the late movement in the polls.

• You Trust The Front-Runner Math: Donalds. A lead in every July poll, Trump's endorsement, a fundraising edge, and a plurality-wins rule in a fragmented field are the loudest facts on the board, and every model made him the clear favorite.
• You Think The Price Is Too Sure: The Collins Gap. The models' most distinctive call is that a sub-penny price on a second-place, DeSantis-appointed sitting lieutenant governor with a statewide ad blitz understates his chance in a race with a big undecided bloc.
• You Want The Honest Bottom Line: This Is Less Settled Than 97 Cents. One rising challenger, a roughly 30 percent undecided bloc, and a two-week runway are exactly the conditions under which a near-certain price carries more risk than it looks.

How To Read These Prices

The models can tell you where they part from the crowd, but the number in front of you still has to be understood in context. Here is how to read a market like this one, plus the honest friction of a thin, event-driven board.

Work an example. Say Donalds's contract shows a 97-cent YES price. Taken at face value, that implies about a 97% chance he wins the nomination. To get a fair, spread-free read you normalize the whole field so every candidate's price sums to 100. On most prediction-market boards the raw prices sum to more than 100 because each contract carries its own bid/ask spread, and de-vigging pulls the favorite's number down. This particular board is unusual: the raw YES prices already sum to about 100 cents, so de-vigging barely moves anything, and Donalds's 97 cents stays about 97%. The lesson holds anyway, always check what a fair version of a price looks like before comparing it to anything, because on a lopsided board the favorite's number is usually softer than it looks. That habit is the whole idea behind converting Kalshi's cents into real odds before comparing anything.

That same discipline of comparing a number to its fair value is what line shopping and finding the +EV side is about. OddsShopper's odds screen and free expert picks cover the sports slate rather than election futures, and the Florida market itself lives on Kalshi, but the method is the same one our Brazil 2026 president model verdict and Venezuela leader model verdict use on very different fields, and every number here lands in our back-tested Model Verdict ledger.

A few honest points before reading too much into any number in the table 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 And Spread. Below the front-runner this board trades thin, most contracts at a single cent or two with light volume, so the "price" is softer than one number suggests, and getting in or out at your number is not always possible.
• A Moving Story. This race has a rising challenger and a large undecided bloc with the primary two weeks out; new events before August 18 can move it, and these estimates reflect only what was known on August 3. Prices shown on this page refresh during the news wave into the primary.
• 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.

FAQ

Who is favored to win the Florida Republican governor primary in 2026? U.S. Rep. Byron Donalds is the clear front-runner. He led every public July 2026 primary poll (ranging from about 37% to 46%), holds President Trump's endorsement plus backing from Sen. Rick Scott and House Speaker Mike Johnson, and has a fundraising edge. Kalshi's market prices him near 97 cents. Our six-model AI panel agrees he leads but is less certain, blending him to about 82%.

What do the AI models think of Jay Collins's chances? The panel gives Lt. Gov. Jay Collins far more chance than the market does, a roughly 14% blend against a market price of under one cent. The models note Collins is the sitting DeSantis-appointed lieutenant governor, sits second in every July poll after a multimillion-dollar ad blitz, and could draw from a large undecided bloc (about 28 to 31 percent in two surveys) in a plurality primary. They still keep Donalds well ahead because his lead held in every poll.

When is the Florida primary and how does the nominee get decided? The Florida primary is Tuesday, August 18, 2026. Florida abolished primary runoffs in 2005, so the nominee is whoever wins a plurality of primary votes, even without an outright majority. In a fragmented multi-candidate field, that means a candidate can win the nomination with well under 50 percent of the vote.

Are Casey DeSantis or Matt Gaetz running for Florida governor? Neither appears as an active candidate in the fetched July 2026 polling. Casey DeSantis publicly weighed a run in 2025 and Matt Gaetz said in January 2025 he was considering one, but neither is listed among the polled candidates in the 2026 Republican field. Their Kalshi contracts remain listed but price near zero.

How does the Kalshi Florida governor nominee market settle? Kalshi lists a separate yes/no contract for each candidate, and each market resolves YES only for whoever wins the Republican nomination for governor at the August 18, 2026 primary, per Kalshi's rules. The board's expected expiration is listed for November 3, 2026, but the nomination itself is decided at the August primary.

How were these model estimates produced? Six AI models scored the entire candidate field without seeing any market prices, working from a fetched, dated summary of the field, the candidacy timeline, and the public July 2026 polls, then read one another's anonymized reasoning and revised. Their per-candidate numbers were blended into one estimate per outcome. These are model estimates, not predictions of fact and not financial or voting advice.

The Bottom Line

The market and the models are looking at the same race and reaching the same front-runner with different conviction. Kalshi prices Byron Donalds near 97 cents, close to settled; the panel blends him to about 82%, still first but a long way from certain, and it gives Jay Collins roughly 14% where the market pays under a cent. The read to take is not that Donalds is in trouble, his lead in every July poll plus Trump's endorsement plus the plurality rule is the biggest fact on the board. It is that a rising challenger, a roughly 30 percent undecided bloc, and a two-week runway leave more unresolved than a near-97-cent price admits. If this method is your thing, our Brazil 2026 president model verdict and our guide to Kalshi versus Polymarket are the natural next reads.

Hottest Prediction Markets Right Now

• 2028 Democratic presidential nominee · $166M traded
• Rocket Classic Winner · $124M traded
• 2027 Pro Football Champion · $51M traded
• Pro Baseball Champion · $40M traded
• Bitcoin price at the end of 2026 · $30M traded
• 2027 Pro Basketball Champion · $19M 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.

Verdicts generated August 3, 2026. Prices as of August 3, 2026. Panel: Claude Fable, Claude Opus, Claude Sonnet, GLM, Kimi, and DeepSeek, price-blind, after one revision round. These are model estimates, not predictions of fact and not financial or voting advice.