Twin Cities Marathon 2026: Testing Our Heat Model on Its History
Twin Cities started at 49°F. We checked our heat model against 24 race days of its history. It tracked well (r = 0.90) and missed 2026 by about a point.
The Twin Cities Marathon started at 8:00 AM on Sunday, October 4, in 49°F air. It was 54°F at 9, 57°F at 10, 60°F at 11 and 62°F at noon. A big field turned out for it: 8,421 official finishers, the most since 2016.
Victor Shitsama of Kenya won the men's race in 2:12:20, his marathon debut. Lilian Chebii, also from Kenya, won the women's race in 2:25:26 and broke the 25-year-old course record of 2:26:51.
Most of this recap is about our heat model, though. Twin Cities has a history of hot early-October races, and we never fit the model on it, so its record makes a fair test. Here's how the model did, including where it missed.
A fair test of the heat model
Our heat numbers come from the RunScore calibration: 6.3 million finish times from 1,686 race days at 92 marathons. Twin Cities is not one of those 92 races. Nothing in its history shaped our heat cost, which makes it a clean out-of-sample check.
Each race day gets two numbers. The predicted cost is what our model charges for that day's weather over the hours the median runner is on the course: heat from WBGT (0.35% per °F above 50°F), plus course wind, plus rain. The actual slowdown is how much slower the field's median finish was than in Twin Cities' own cool years, after we remove the race's baseline speed, a year-to-year effect pooled across all races, and Twin Cities' own long-run trend. If the model is any good, the two should line up. (If WBGT is new to you, we cover it in our WBGT explainer.)
Across 24 race days from 2001 through 2026, they do. The correlation between predicted cost and actual slowdown is r = 0.90. There was no race in 2020 or 2023, and 2000 has no weather record, so those years are out.
The hot years
A correlation can hide big misses on the days that matter, so here are the three hottest race days in the record:
| Year | WBGT | Predicted slowdown | Actual slowdown |
|---|---|---|---|
| 2007 | 72°F | +8.1% | +10.9% |
| 2025 | 64°F | +5.8% | +5.6% |
| 2005 | 63°F | +5.1% | +3.9% |
2025 came out nearly exact. 2007 ran almost three points slower than the model said, and 2005 about a point faster. The model gets the order and rough size right, and it isn't precise on any single day.
There is also a plain outlier. In 2006 we predicted +2.0% and the field ran +5.5% slower. We don't have an explanation for that year.
2026: a miss of about a point
The average WBGT over the median runner's hours was 52.6°F, barely over the 50°F line where our heat cost starts. We predicted the field would run about 1.1% slower than in a cool year. The actual figure was 2.2% slower. That's a miss of about one point, in the slower direction.
So where did the extra time come from?
Where the extra time came from
The cleanest comparison is against 2024, another cool year. Here is the finish time at each point in the field, net time:
| Point in the field | 2024 | 2026 | Change |
|---|---|---|---|
| 10th percentile | 3:23 | 3:25 | +0.9% |
| 25th percentile | 3:48 | 3:50 | +1.2% |
| Median | 4:18 | 4:22 | +1.8% |
| 75th percentile | 4:50 | 5:01 | +3.7% |
| 90th percentile | 5:23 | 5:40 | +5.1% |
The front of the field barely moved. The back slowed by a lot more, and the field was bigger: 8,421 finishers in 2026 against 6,781 in 2024, about 1,640 more.
Our read is that the slowdown sits at the back of a field that grew, and that on a morning that started at 49°F, heat doesn't explain it. That's our interpretation. This data can't prove who the extra finishers were or why they were slower.
Comparing a race's median against its own history counts any change in who runs as a miss. A larger, slower-leaning field looks the same in the numbers as a hotter day would.
Why the race moves in 2027
Twin Cities In Motion is moving the marathon to mid-October starting in 2027, citing recurring early-October heat. The 2023 race was cancelled for heat (Black Flag conditions). Among the race days we can price, the hot ones cost this field a lot: the actual slowdown was 3.9% in 2005, 5.6% in 2025 and 10.9% in 2007.
Sources
Results, winners and the finisher count come from the official race results, with winner details cross-checked against CBS Minnesota and MPR. Finisher counts by year come from the race's historical results page. Race-day weather is Racecast's observed hourly record for October 4, 2026, with MPR reporting about 50°F at the start and 57°F by 10 AM. The predicted and actual slowdown figures come from the same audit as our RunScore calibration.
More guides
- Why We Rebuilt RunScore From 6.3 Million Marathon Finishes: where the heat model and its calibration come from
- WBGT Explained: The Number That Decides If Your Race Gets Cancelled: the heat-stress measure used above
- How to Run a Marathon in the Heat: pacing, hydration, and gear for a warm race day