The 7 Most Contested Prediction Markets On Kalshi Right Now

Updated August 10, 2026 · 14 min read · by Jake Hari

Every week our panel of AI models prices dozens of Kalshi markets blind, no prices shown, working only from fetched data. Most of the time the eight seats land within a few points of each other, and the interesting number is the gap between the panel and the market. This column is about the other case: the markets where the models fight each other. A market where eight independent readers of the same evidence land 45 points apart is telling you something no single price can, which is that the evidence itself does not pin the answer down. Welcome to Edition 1 of the Disagreement Index, a weekly ranking of the open markets our panel can't agree on.

The Quick Answer

Out of roughly 1,400 market runs our panel scored in the past week, these seven open markets carry the widest seat-to-seat splits: NASA's crewed Moon landing by 2029 (seats from 30% to 75% against a 48-cent market), Bronny James still a Laker on opening night (56% to 88%), Tyreek Hill signing nowhere or retiring (48% to 77%), Jacob Misiorowski winning NL Cy Young (30% to 59% against an 84-cent market), Democrats winning the House (69% to 97.5% before the revision round collapsed the split to 3 points), a 2027 U.S. recession (11% to 33%, and not one seat moved an inch after debate), and the Ohio Senate race (40% to 61%, where one seat read the arguments and crossed to the other side). The full board, every seat's number, and what the models actually argued are below.

Free: The Weekly PM Market Brief — the AI panel's graded record, the week's biggest market-vs-model gaps, and what's spiking next. One email, Sundays. The signup box is at the bottom of this page, or keep reading: every split on this board comes with the reasoning behind both poles.

A disagreement is not a pick. When the panel splits, the honest reading is that the market's confident-looking price may be papering over real uncertainty, sometimes about the event, sometimes about the rules that settle it. Our working hypothesis, which this column will test in public over many editions, is that high disagreement flags ambiguous premises and hidden information. That is a hypothesis, not a validated law, and we will print the correlation either way once enough of these markets settle. New to event markets? Start with how Kalshi contracts actually work.

How We Measure Disagreement

The pipeline behind this board is the same one that powers our graded verdict pages. Each market is scored price-blind by up to eight model seats (three Claude models, GPT, Gemini, Kimi, GLM, and DeepSeek, with Grok joining some runs), each working from the same fetched data card. Round 1 numbers are recorded, then every seat reads the others' anonymized reasoning and may revise; round 2 is stored separately, never overwritten. Disagreement is measured on round 1, where the spread is max minus min across seats. The revision round then tells us which kind of disagreement we caught: a split that converges was mostly seats weighing the same evidence differently until they compared notes; a split that holds means the seats have irreconcilable readings, and that is the interesting signal. Runs in this edition were generated August 3-9, 2026; every market below was re-verified as open, with live prices, on August 10, and nothing here settles inside the next 48 hours. Not every run included all eight seats, and two of the seven boards ran without a revision round; both facts are marked where they apply.

The Board

Market	Kalshi Price	Round 1 Split	After Revision	Panel Blend
NASA Moon landing by 2029	48¢	45 pts (30-75)	27 pts, held	55.9%
Bronny James: still a Laker	83¢	31.6 pts (56-88)	no revision round	73.4%
Tyreek Hill: no new team / retires	39¢	29.3 pts (48-77)	no revision round	61.8%
Misiorowski wins NL Cy Young	84¢	29.2 pts (30-59)	19.1 pts, mostly held	42.3%
Democrats win the House	85¢	28.5 pts (69-97.5)	3 pts, converged	85.1%
U.S. recession in 2027	25¢	22 pts (11-33)	22 pts, frozen	18.9%
Ohio Senate: Sherrod Brown	48¢	21 pts (40-61)	12 pts, narrowed	49.9%

Kalshi prices are the most recent traded prints as of August 10, 2026. "Panel Blend" is the equal-weight mean of the final round that ran (round 2 where one exists, round 1 otherwise). Split figures are round-1 max minus min in probability points.

More live boards from the same panel: Kalshi longshot picks · Sanchez 8¢ · Kalshi Fine Print Watch · the 10 biggest markets graded A-F · 2028 Democratic nominee board · Newsom 17¢

Every Seat's Number

Round 1 estimates, before any seat saw the others' reasoning. A dash means that seat did not run that board.

Market	Kalshi	Fable	Opus	Sonnet	GPT	Gemini	Kimi	GLM	DeepSeek	Grok
Moon by 2029	48¢	50	30	58	61	45	55	70	75	—
Bronny stays a Laker	83¢	84.8	87	72.9	72	88	56.4	58.2	68	—
Hill: no new team	39¢	47.5	76.8	57	67	—	55	67.4	62	—
Misiorowski NL Cy	84¢	47.4	29.8	32	54.4	59	51.5	41.4	30	—
Democrats win House	85¢	81	84.5	69	85	—	83	85	97.5	—
Recession in 2027	25¢	11	11	18	20	20	14	20	33	—
Ohio Senate: Brown	48¢	55	40	55	54	61	46	—	—	54

These are model estimates, not predictions of fact and not financial advice, and on the political markets this page takes no side in any race; it analyzes disagreement about prices. Kalshi is a CFTC-regulated event-contract exchange open to eligible U.S. users 18+ (18 and older), and availability varies by state as of August 2026.

Every seat on this panel is graded against real market settlements — records to date: Claude Fable 82% on 450 graded calls · Claude Opus 84% on 514 graded calls · Claude Sonnet 82% on 498 graded calls · GPT 83% on 3,225 graded calls · Gemini 84% on 3,252 graded calls · Kimi 83% on 2,285 graded calls · GLM 81% on 2,307 graded calls · DeepSeek 80% on 2,332 graded calls · Grok 82% on 2,865 graded calls. Recomputed daily; the full scoreboard is public.

No. 1: The Moon Deadline, Where The Panel Split 45 Points And Stayed Split

The widest live split in our window is NASA landing a crewed mission on the Moon before 2030, a market trading at 48 cents. In round 1 the seats ranged from Opus at 30% to DeepSeek at 75%. Every seat describes the same mechanism and then weighs it differently. Fable, at 50%: "A serial chain — HLS refueling demo, Artemis 3 LEO demo, then Artemis 4 — must hold with under a year of aggregate slip; this program has slipped four years in five." Opus, the low pole: "A crewed landing sits behind a serial chain (LEO demo, uncrewed HLS with orbital refueling) whose first links haven't flown, and this program's dates slip 12-24 months routinely." DeepSeek, the high pole, read the same chain as survivable: "first-of-kind refueling likely slips landing into 2029 but serial testing reduces catastrophic risk."

Then came the revision round, and this is what makes the Moon the flagship of Edition 1: the split narrowed only from 45 points to 27 and the poles refused to close. DeepSeek came down from 75 to 65 and stayed the high pole; Opus moved just 30 to 38, writing that the in-year buffer argument was real "but the un-demonstrated cryogenic transfer campaign plus two serial demos still makes a clean chain the minority case." Kimi did not move at all: "Holding my ladder: the panel's new arguments mostly restate my own mechanism," it wrote, adding that each of the two opposing poles "overlook a fact I already priced." Eight readers, the same fetched program history, a 27-point standing disagreement. The companion market, a landing by the end of 2028, trades at 22 cents and drew the same shape of fight (spread 38 in round 1, 21 after revision). When a split survives structured debate, the uncertainty is in the world, not in the readers.

No. 2 And No. 3: Two Roster Questions With No Revision Round

Bronny James, 83 cents to still be a Laker when next season opens. Seats ran from Kimi at 56.4% to Gemini at 88%. The high seats treat the incumbency as settled: Gemini notes the Lakers "held James out of Summer League citing his secure roster spot," and Opus points to the $2.3M salary the Lakers locked in fully back in June. The low seats accept every one of those facts and still discount the outcome on the roster math, with Kimi calling retention only "the modal outcome despite the 16-to-15 crunch." Nobody disputes the evidence; they dispute how much a locked-in contract is worth when someone has to be the odd man out.

Tyreek Hill, 39 cents on the "stays with Miami or retires" outcome of his next-team board, which in practice is the market's bucket for "signs with no new team in 2026." Opus, at 76.8%: "A 32-year-old ten months from a multi-ligament knee dislocation with no suitors and no medical clearance most likely signs nowhere in 2026." Fable, the low seat at 47.5%, reads the same file with more room for a late signing, noting "Kansas City is the only card-verified signing story" if he is cleared. A 29-point split on a question with a hard December deadline.

Both runs were single-round, so we cannot say whether these splits would have held through debate. That disclosure matters, and future editions will prefer boards with a stored revision round.

No. 4: Misiorowski, Where The Panel Fights The Market And Itself

Jacob Misiorowski for NL Cy Young trades at 84 cents, and this row is doing double duty: the seats split 29 points in round 1 (Opus 29.8 to Gemini 59), and even the highest seat sits well below the market. Gemini's bull case: he "combines historically elite rate stats with a 180-inning pace that safely clears minimum voting thresholds, making him the overwhelming favorite absent a late-season injury." Opus's discount is about epistemics, not the pitcher, and its revision-round note is worth quoting at length: "I stay meaningfully below the 47-59% cluster because the discount is about what this card cannot see, not about the pitcher. Thirty percent of the season is unplayed, and the card explicitly excludes injury news." The revision round narrowed the split to 19 points and the blend barely moved, from 43.2% to 42.3%. Half the seats treat an unfinished season as a reason for humility, half treat the statistical gap to the field as decisive, and an 84-cent market is quietly siding with one camp. The runner-up mechanics the seats keep naming, Cristopher Sanchez's workhorse profile, are the same ones on our Kalshi longshot board.

No. 5: The House, Where Debate Actually Worked

Democrats to win the House trades at 85 cents. Round 1 looked like a real fight: Sonnet at 69%, DeepSeek at 97.5%, five seats clustered in the low-to-mid 80s. Then the revision round did exactly what it is designed to do. Sonnet came up 14 points after crediting the polling-environment evidence it had discounted; DeepSeek came down 12.5 after peers pushed back on near-certainty; the final round spans 83 to 86, a 3-point band, blend 85.1%, sitting almost exactly on the market. Both poles wrote down what moved them, in nearly mirror-image language about the same two facts, a D+6.7 generic ballot against a Senate map that constrains the companion chamber markets. This is the converged case: the round-1 spread was measurement noise in how seats weighted shared evidence, not a disagreement about the world, and one structured exchange of reasoning dissolved it. Compare that with the Moon market above, where the same procedure barely dented the split. The distinction between those two outcomes is the entire reason round 2 exists. As with every political market we cover, the panel prices probabilities from fetched polling data and takes no side; the House market's own settlement mechanics, which have a fine-print wrinkle worth knowing about, got a C+ in this week's Fine Print Watch.

No. 6: The 2027 Recession, Where Nobody Moved An Inch

This one is our favorite, and not because of the size of the split. A U.S. recession in 2027 drew seats from 11% (Fable, Opus) to 33% (DeepSeek) in round 1, a 22-point spread. After the revision round, the spread was still exactly 22 points, and the blend had moved from 18.4% to 18.9%. Reading the stored revision notes, both poles explicitly named the other side's best argument and rejected it. DeepSeek, holding 33%: the strongest opposing argument, that tight credit, low claims, and Fed easing make a 2027 recession unlikely, "is a pattern frequently observed in the quarters just before past recessions." Fable, holding 11%: the un-inversion danger-window argument "was already central to my round 1 read; no peer added corroborating stress evidence (credit, claims, policy) that the mechanism needs to actually fire." Kimi split the difference and moved a grand total of 2 points, crediting the observation that "early 2007 showed identical calm" in the coincident data. This is a frozen split: full information exchange, zero convergence. Worth noting honestly: when this run was scored on August 3 the market traded at 40 cents; it has since drifted to 25, toward the panel's blend. We flag that as an observation with a one-week sample, nothing more.

No. 7: Ohio, Where A Seat Crossed The Aisle

The Ohio Senate special election is the closest-priced market on this board, 48 cents on Sherrod Brown against 51 on Jon Husted, and the panel is just as tight: seven seats from 40% to 61% on Brown, blending to 52.1% in round 1 and 49.9% after revision. What earns Ohio its slot is the single most interesting revision arc of the week. Gemini opened as Brown's best seat at 61%, arguing "strong Democratic midterm fundamentals and Brown's proven state overperformance offset Ohio's Republican baseline." After reading the other seats, it did not trim its number; it crossed the midline entirely, landing at 42% and writing that "Ohio's deep structural Republican lean and historically D-skewed polling give Husted a clear, albeit narrow, advantage." The specific fact that moved it appears in Opus's round-1 note: "Ohio polling has understated Republicans by ~4 points in both recent Senate races." A seat abandoning its own side of a coin-flip because a peer surfaced a systematic polling error is the revision round earning its keep, whatever November brings. The panel's final read: a true toss-up, in a market priced like one.

When Disagreement Is Just A Misread

One more market posted a bigger raw split than anything on this board, and we are deliberately not ranking it, because it is the cautionary tale of the method. Tim Cook leaving Apple in 2026, now a 99-cent market with a board-approved September 1 transition, drew seven seats between 97.5% and 99.3% in round 1, and DeepSeek at 0.2%. A 99-point spread! Except DeepSeek's written rationale argued the same case as everyone else, that "reversals of an officially dated megacap CEO succession occur at <0.5% base rate." It had priced the complement: its number answered the mirror-image question its own reasoning ruled out. In the revision round it flipped to 99.8%, and the panel's spread collapsed to 1.2 points. The lesson, and the reason this column reports round-2 behavior and not just raw spreads: a huge disagreement number can mean deep uncertainty about the world, or it can mean one reader held the map upside down. The revision round is how we tell those apart, and Fable's Apple note states the residual plainly: "the panel's spread reflects tail-risk taste, not new information."

Settlement Timeline

Every market on this board settles on a fixed public schedule, and several resolve Yes or No well before their final date if the event fires early.

Market	Settles	Sooner Trigger
Moon by 2029	End of 2029	Any Artemis schedule announcement
Bronny James	October 23, 2026	Roster cutdowns
Tyreek Hill	December 31, 2026	Medical clearance news
Misiorowski NL Cy	November 2026 (award)	September innings decisions
House control	February 2027	Election night, November 2026
Recession 2027	January 2028	Each claims/credit print
Ohio Senate	January 2027	Election night, November 2026

Every number in this piece is stored with its run date in our backtest ledger, and every market above gets graded against the panel's blend and each seat's number when it settles; disagreement-vs-outcome is a measurable question and this column exists to measure it in public on the graded scoreboard. Prices refresh as these markets move, and each edition of this page re-scores the last one on the same URL.

Hottest Prediction Markets Right Now

• 2028 Democratic presidential nominee · $176M traded
• 2028 U.S. Presidential Election winner? · $58M traded
• 2027 Pro Football Champion · $58M traded
• 2028 Republican presidential nominee · $56M traded
• Pro Baseball Champion · $50M traded
• More tech layoffs in 20​26 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 Bottom Line

Seven markets where eight readers of the same evidence could not agree, one where debate dissolved the fight, one where nobody budged, and one where the scariest-looking split turned out to be a misread. If you trade any of these, the takeaway is not a direction; it is that the single confident number on the screen is hiding a fight that even structured, evidence-sharing debate could not settle. Markets where the panel splits and holds its split deserve a second reading of the rules and a smaller size than conviction markets, whichever side you like. Edition 2 runs next week: same method, fresh runs, and the first entries in the public record of what panel disagreement actually predicts.