Tomorrow's NYC High, Priced: 8 AI Models vs The Weather Crowd
By Jake Hari
July 27, 2026
TL;DR
The panel's most likely bands: 81° to 82° (31%), 83° to 84° (29%), 85° or above (16%). Each model priced the entire band set in one coherent pass: every column sums to 100 — from live data fetched at generation time.
This is the fastest test on our scoreboard: Kalshi's board on tomorrow's Central Park high settles within 36 hours of publish. The models got the live NWS observations and hourly forecast: the same public data the settlement uses — making weather the purest model-vs-crowd arena we cover, and the category where our graded record shows models performing closest to the market.
The Bands
| Band | Kalshi | AI blend | ChatGPT (GPT-5.5) | Claude Fable | Claude Opus | Claude Sonnet | Gemini 3.1 Pro | GLM 5.2 | Kimi K3 | DeepSeek V4 |
|---|---|---|---|---|---|---|---|---|---|---|
| 77° to 78° | 35¢ | 6% | 7% | 4% | 5% | 7% | 7% | 5% | 7% | 5% |
| 79° to 80° | 34¢ | 15% | 26% | 13% | 12% | 15% | 15% | 10% | 15% | 15% |
| 76° or below | 20¢ | 3% | 1% | 2% | 3% | 3% | 3% | 5% | 4% | 3% |
| 81° to 82° | 14¢ | 31% | 42% | 30% | 22% | 32% | 25% | 30% | 26% | 40% |
| 85° or above | 5¢ | 16% | 4% | 18% | 30% | 13% | 25% | 15% | 20% | 7% |
| 83° to 84° | 3¢ | 29% | 20% | 33% | 28% | 30% | 25% | 35% | 28% | 30% |
An honest note before the table's grade comes in: the panel's top bands sit a few degrees above where the crowd concentrates, and the crowd's read tracks the official forecast more closely. Thirty-six hours from now the observation settles who was right, and the result goes on the scoreboard either way.
The Reasoning
ChatGPT (GPT-5.5): For July 28 at KNYC, I center the forecast near 82F: the live NWS guidance shows today already topping out in the low/mid 80s, then the Tuesday outlook is cooler and wetter with showers/thunderstorms likely and a high near 82F. Rain/cloud timing is the main uncertainty. If showers thicken by late morning, the high could stall in the upper 70s or around 80F; if convection holds off until mid/late afternoon, Central Park can still reach 83-84F. I put limited probability on 85F+ because the synoptic setup appears cloudier/wetter than today, but July urban heating and forecast error keep it possible. A 76F-or-below high would require unusually early, persistent rain and poor mixing, so it is very unlikely.
Claude Fable: The live card is dominated by July 27 data: observed max 84F today, NWS high near 83 with temps easing to 81 amid scattered thunderstorms, tonight's low around 70 with continued shower chances. No explicit hourly forecast for July 28 is provided, so tomorrow's distribution is anchored on persistence of this warm, moist late-July pattern: a 70F overnight low with south flow typically rebounds to the low-to-mid 80s at Central Park. Central mass goes to 81-82 and 83-84, roughly matching today's 83-84 outcome. Cloud cover and convection could cap the high near 80, so 79-80 gets meaningful weight; a break in clouds could push 85+, which gets a real but smaller tail. Sub-78 outcomes would require a washout or unadvertised frontal passage and are priced thin.
Claude Opus: The live card describes today (7/27), not the settlement day, so it constrains tomorrow only through persistence. Today reached 84F with a forecast high near 83 and afternoon cooling, under partly sunny skies with scattered convection and a light south wind. Tonight's low near 70 signals a warm, humid airmass staying in place rather than a frontal push. Late-July Central Park climatology averages roughly 85F for the daily max, and day-over-day maxima in this pattern typically vary by only 2-4F absent a front. The chief downside risk is persistent mostly cloudy skies and morning/midday showers capping the diurnal rise, which could hold the high in the low 80s or upper 70s. The chief upside is a break in the clouds with continued southerly warm advection, pushing 85-87F. I therefore center the distribution on 83-84 and 85+, with a meaningful 81-82 tail and thin probability below 80F.
Model estimates generated July 27, 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 exchange; 18+, availability varies by state.
FAQ
Why do the model percentages differ from the Kalshi price?
The models never see the price. When they disagree with the crowd, one side is wrong, and we grade every verdict against real settlements on our public scoreboard.
Are model verdicts betting advice?
No. Model verdicts are model estimates, not betting or financial advice. Treat them as one input among many and make your own decisions.