TL;DR
- Kalshi's 15 threshold markets price tonight's speech at a ~30-minute median - and seven blind AI models independently landed on the same number.
- The panel flagged '34+ minutes at 46¢' as the board's most overpriced rung on Wednesday; by this morning the market had marked it to 33¢ - the models' exact number.
- What's left: 28+ minutes at 61¢ still trades above the 53% blend, and the panel's long-speech dissent (Fable, median 34) is the case to watch if the night runs triumphant.
Kalshi doesn't just ask what President Trump will say at tonight's White House Correspondents' Dinner: a 15-market ladder asks how long he'll say it, one threshold at a time, from "12 or above minutes" (98¢) to "40 or above" (19¢). We had seven AI models study his actual speaking record and price the same ladder on Wednesday: each model building one probability distribution for the speech's length and reading all 15 thresholds off it, without ever seeing a market price. Then something happened that's worth an article on its own: overnight, as the program details firmed up, the market repriced the whole ladder directly onto the models' curve.
The Market
| Venue | Kalshi — a CFTC-regulated exchange (18+; availability varies by state, as of July 2026) |
| The event | "How long will Trump speak for at the White House Correspondents' Dinner?" — 15 yes/no contracts, one per threshold |
| The contract | Each market resolves YES if Trump speaks for the stated number of minutes or more at the July 24, 2026 dinner, per Kalshi's settlement rules |
| Event date | Tonight — Friday, July 24, 2026, 7:00pm EDT |
| Prices below as of | the morning of July 24, 2026 |
The Setup: A Format That Says 20, a Speaker Who Says 60
Duration markets are a tug-of-war between the format and the speaker, and rarely has that rope been pulled harder from both ends:
The format says short. A president's correspondents'-dinner set is traditionally a tight comedic routine inside a crowded program: historically 10 to 30 minutes, not a solo address. Friday's program is fuller than most: the rescheduled dinner will honor the Secret Service agents and hotel staff from April's attack, alongside the traditional awards and scholarships, all under heightened security at a reduced-capacity Waldorf Astoria.
The speaker says long. Trump's February State of the Union ran 1 hour, 47 minutes, 40 seconds: the longest address to a joint session of Congress in American history, and his second consecutive record. And this isn't stale history: at a "Remarks" event in Marietta, Georgia the day before this dinner, he spoke for over 60 minutes. That is not an estimate; it is the settled result of Kalshi's own duration ladder for that event.
And this speech has been waiting since April. The original dinner was cut short by gunfire moments after the WHCA president welcomed Trump to the ballroom; he never delivered his planned remarks. He returns with three months of material and a survival story. And per this morning's reporting, the program is fuller still: mentalist Oz Pearlman returns as the entertainer, Trump is expected to "briefly address" the April shooting, and he plans to reuse jokes originally written for that night.
Our models received exactly this context of dated, sourced facts. Because all 15 thresholds describe one speech, each model was required to price the entire ladder from a single internal estimate of the length distribution, so its 15 probabilities are internally consistent by construction. And as always in this series, the models never saw the market prices.
The Ladder: 15 Thresholds, Seven Models
| Threshold | Market | Fable | Opus | Sonnet | GPT | GLM | Kimi | DeepSeek | AI blend | Edge |
|---|---|---|---|---|---|---|---|---|---|---|
| 12+ min | 98¢ | 97 | 96 | 95 | 96 | 96 | 95 | 99 | 96% | −2 |
| 14+ min | 93¢ | 96 | 94 | 91 | 93 | 94 | 91 | 98 | 94% | +1 |
| 16+ min | 91¢ | 94 | 91 | 87 | 89 | 90 | 86 | 95 | 90% | −1 |
| 18+ min | 84¢ | 92 | 87 | 81 | 84 | 85 | 81 | 91 | 86% | +2 |
| 20+ min | 79¢ | 89 | 82 | 75 | 78 | 79 | 76 | 86 | 81% | +2 |
| 22+ min | 74¢ | 85 | 75 | 68 | 72 | 71 | 71 | 79 | 74% | 0 |
| 24+ min | 70¢ | 80 | 68 | 60 | 66 | 63 | 65 | 71 | 68% | −2 |
| 26+ min | 61¢ | 74 | 60 | 52 | 61 | 54 | 59 | 61 | 60% | −1 |
| 28+ min | 61¢ | 68 | 55 | 44 | 56 | 45 | 54 | 51 | 53% | −8 |
| 30+ min | 48¢ | 62 | 50 | 37 | 51 | 36 | 49 | 41 | 47% | −1 |
| 32+ min | 41¢ | 55 | 43 | 30 | 45 | 28 | 44 | 31 | 39% | −2 |
| 34+ min | 33¢ | 49 | 37 | 24 | 39 | 21 | 39 | 23 | 33% | 0 |
| 36+ min | 25¢ | 43 | 31 | 19 | 33 | 15 | 34 | 16 | 27% | +2 |
| 38+ min | 21¢ | 38 | 26 | 15 | 27 | 10 | 30 | 10 | 22% | +1 |
| 40+ min | 19¢ | 33 | 22 | 11 | 22 | 7 | 26 | 7 | 18% | −1 |
Model estimates generated July 23, 2026, price-blind, each model pricing the full ladder from one distribution. 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.
The One-Number Read
Reduce each column to the speech length where its probabilities cross 50%, and the picture is striking for how close it is:
- The market's implied median: about 30 minutes (48¢ at 30+, on the morning-of-event board).
- The models' implied medians: 26 to 34 minutes, centering right on 30. Five of seven (Opus, ChatGPT, Kimi, DeepSeek at 30; Sonnet at 27) sit within a few minutes of the market; GLM is the short end at 26; Fable the long end at 34.
In other words: the crowd and the panel independently agree this is a roughly half-hour speech: the format's gravity and the speaker's overrun habit approximately cancel. DeepSeek's rationale captures the consensus: "Trump's relentless verbosity and record-breaking SOTU length collide with a structured dinner format", and the collision lands near 30. When seven blind models confirm a price rather than fight it, that's worth knowing too: it says the market isn't sleepwalking here.
The Overnight Convergence
Here is the ladder the models were actually scored against. On Wednesday, when the panel's verdicts were generated, the market said 46¢ at 34+ minutes and 34¢ at 36+, and the panel scored those rungs 33% and 27%, flagging "34+ at 46¢" as the most overpriced contract on the board. By this morning, with the full program confirmed (Pearlman's set, the Secret Service tribute, the First Amendment segment), the market had marked 34+ down thirteen cents to 33¢. The panel's number exactly. The move wasn't one rung: 30+ went 52→48, 32+ went 47→41, 36+ went 34→25. Rung by rung, the crowd walked onto the curve seven blind models had drawn a day earlier.
We say this with the required humility: one converging ladder is one data point, the models were built from public facts any sharp trader also had, and tonight the speech itself will grade everyone, the panel included. But this is the cleanest live demonstration yet of what this series is for: price-blind model estimates that anticipated where the market went next.
What's Left on the Board
After the overnight move, the ladder is, by the panel's lights, almost perfectly priced. Thirteen of fifteen rungs now sit within two points of the blend. Two small residuals:
- 28+ minutes at 61¢ (blend 53%) is the one rung still carrying a premium. The market's curve barely drops from 26+ (61¢) to 28+ (61¢), while the models' distribution steps down five points there. If anything on this board is still rich, it's this.
- The 20-minute floor at 79¢ (blend 81%) is the mildest of leans the other way; the models see the "brief somber set" scenario as slightly less likely than the market does, given the three months of saved-up material.
Neither is the kind of gap we'd write a headline about. The honest summary is better than a forced pick: the market and the panel now agree this is a ~30-minute speech, and the remaining disagreements are pocket change. Fable, the panel's long-speech outlier at a 34-minute median, with a third of its probability still on 40+, is the one dissent worth remembering if the night runs triumphant instead of solemn: "Trump's documented 2-3x overrun pattern (107-min SOTU, 60+ min the day before) dominates the 10-30 minute WHCD norm."
To be explicit: these are model estimates for information and entertainment, not financial advice and not a recommendation to trade any contract. Our models are graded against real settlements all summer and are wrong regularly. Event contracts are for adults 18+ where legal. For the rest of tonight's picture, see the companion pieces on the 34 mention markets and the reschedule market itself.
FAQ
Is Kalshi legal in my state?
Availability varies by state and changes over time, so check your eligibility directly on the platform before doing anything. Kalshi is a federally regulated exchange (event contracts are CFTC-regulated and are not sports betting), but access rules differ by jurisdiction and you must be 18 or older. That is the state of play as of July 2026.
Are the model verdicts advice?
No. Model verdicts are statistical estimates of how a market might resolve, not financial advice, and nothing here is a recommendation to buy or sell any contract.
How is speaking time measured?
Per Kalshi's settlement rules for each contract: the market resolves YES if Trump speaks for the stated number of minutes or above at the July 24, 2026 dinner. Kalshi's rules govern what counts as the start, end, and interruptions of the remarks; read the contract for edge cases before assuming.
Why price the whole ladder at once instead of each market separately?
Because all 15 contracts describe one speech. Pricing them independently can produce probabilities that contradict each other (a higher chance of clearing 40 minutes than 38, say). Each model in this article built one probability distribution for the speech's length and read every threshold off it, so its ladder is internally consistent by construction.
Why didn't the models see the market prices?
Anchoring. Our summer-long grading shows that models shown the market price cluster tightly around it, which makes their estimate an echo rather than an opinion. Price-blind estimates are what make the ladder gaps meaningful, and also riskier: sometimes the crowd knows something the models don't. On this ladder, the blind panel largely confirmed the crowd, which tells you the disagreement that remains is specific, not reflexive.
Will these prices move before Friday?
They already did: the entire ladder repriced overnight as the program details firmed up. The table above is the morning-of-event snapshot, July 24; expect more movement right up to the podium.
