Ben Dhiman won the 2026 UTMB in 18 hours, 16 minutes and 29 seconds — the fastest winning time ever recorded at the race and the first finish inside 19 hours. UTMB lists the 2026 course at 174 km with 9,700 m of climbing. The start was delayed two hours for storms and organisers removed the Pyramides Calcaires section, so the time should not be read like a road-race course record.
The more useful comparison is with himself. Dhiman finished second at UTMB in 2025 in 19:51:37, on a course UTMB lists at 174.2 km and 9,920 m — near-identical, and also weather-modified, with the same Pyramides Calcaires omission. Same runner, same race, essentially the same route: 1 hour 35 minutes faster, and second place to first.
He was not the only one moving quickly. Baptiste Chassagne finished second in 18:47:38, Caleb Olson third in 18:48:24 just 46 seconds later, and Virgile Moriset fourth in 19:00:23. Three men broke 19 hours on the same day and a fourth missed it by 23 seconds. Something about this edition was fast. The results establish that; they do not tell us why.
But the number that impresses me most is not 18:16.
It is 31:09 — Dhiman’s winning margin. He moved to the front around La Balme, 41 km in, with Chassagne still only about 45 seconds back at the Croix du Bonhomme, then made the decisive break on the descent toward Les Chapieux and never gave the lead up.
The advantage was not perfectly linear, and that makes the race more interesting. Chassagne reduced the gap from 28:12 at Champex-Lac to 25:53 at Trient. Dhiman then opened it again, reaching Vallorcine 30:47 clear. He did not disappear early and cruise home. He built a large lead, saw part of it come back, and extended it again late.
There is also a recorded GPS analysis of his race, broken into 19 segments: 169.38 km, +9,843 m, 18:14:01, average cadence 152 spm. It is not an official measurement — the data caveats are collected near the end of this piece, and they matter — but for comparing one part of his race against another, it is genuinely useful.
One limit is worth stating up front, because it shapes everything below. The trace shows how Dhiman produced his performance. The official checkpoint splits show where he separated from the field. What no public dataset gives us is equivalent terrain-by-terrain movement data for Chassagne or Olson. So when the trace says he climbed a section at 822 vertical metres an hour, we cannot say that climb won him the race. We do not have the denominator.
One race, many different jobs
The platform reports an average pace of 6:29/km, and that number tells us remarkably little.
Across the 19 segments, Dhiman moved from 4:17/km on runnable ground to 11:28/km on the steepest climbing.
The opening 8.77 km from Chamonix to Les Houches took 37:33 — 4:17/km, despite containing 234 m of climbing and 262 m of descent. Then the job changed. From Les Houches to Col de Voza he covered 5.25 km with 680 m of ascent in 37:57. Pace fell to 7:14/km, cadence from 171 to 140 steps per minute, and the segment produced a vertical gain rate of about 1,075 m/h.
Nothing had gone wrong. The terrain required a different form of locomotion.
A 4:20 kilometre on runnable trail and a 12-minute kilometre up a steep Alpine climb are both recorded as one kilometre. Mechanically, they are different jobs.
The full-race gradient classification makes the point at scale:
| Gradient | Distance | Share |
|---|---|---|
| Above 20% up | 15.81 km | 9.3% |
| 15–20% up | 11.26 km | 6.6% |
| 10–15% up | 14.04 km | 8.3% |
| 5–10% up | 17.53 km | 10.3% |
| −5% to +5% | 50.94 km | 30.1% |
| 5–15% down | 33.17 km | 19.6% |
| Steeper than 15% down | 26.63 km | 15.7% |
Only 30% of the recorded distance was anything close to flat. About a quarter was uphill at 10% or steeper, more than a third downhill steeper than 5%. And 4,257 m of ascent — some 43% of all the climbing in the trace — came on ground above 20%.
This was not a running race interrupted by hills. The hills were the race.
The biggest climb
The largest single ascent came from Les Chapieux to Col de la Seigne: 10.51 km, +1,065 m, 1:17:45, at 822 vertical metres an hour.
Duration matters here. A short steep climb produces a high vertical rate easily, because it only has to be sustained briefly; comparing a 20-minute rate with an 80-minute one is not like-for-like. Dhiman accumulated more than a vertical kilometre while holding above 800 m/h for well over an hour, in darkness, before halfway.
The number is interesting not because 822 is spectacular, but because of how much climbing it covers and how long he held it.
The steep climbs
Two later sections stand out for gradient rather than size.
Courmayeur to Refuge Bertone — 5.28 km, +802 m, 51:19, 937 m/h, with around 31% of the segment classified above 20%.
Arnouvaz to Grand Col Ferret — 4.37 km, +762 m, 50:18, 909 m/h, roughly 45% above 20%, and individual kilometres at 20.7% and 24.3%.
On this ground the useful distinction is no longer fast running versus slow running. It is running versus power hiking.
His cadence on Grand Col Ferret averaged 120 spm, the lowest of the 19 segments. On the descent immediately after — 9.7 km and 1,025 m down to La Fouly — it returned to 171.
Cadence does not tell us how hard either effort was, or whether economy and force production had deteriorated underneath. But the abrupt change is useful descriptively: he had not become trapped in one slow movement pattern. The terrain changed and so did he.
That switching is more interesting than the cadence itself. Strong mountain runners can climb quickly. Ultra performance asks the harder question — can you climb hard, descend afterwards without destroying the legs, and then run again when the trail allows it?
By Grand Col Ferret his lead was already around 22 minutes.
The climb that matters more
The most impressive number may be one of the smaller ones.
Late in the race came Vallorcine to La Flégère: 10.97 km, +961 m, 1:33:39. Dividing ascent by total segment time gives 616 m/h — a whole-segment descriptor, not his climbing speed, since the section includes flatter and descending ground.
Beside the earlier 800s and 900s, 616 looks ordinary. But look at when it happened. Dhiman had covered more than 150 km and raced through the night, and around 10 km from the finish, partway up this climb, he tripped on a rock and fell hard. He got up, completed the ascent to La Flégère, and carried on toward Chamonix.
The number itself is unremarkable. Its position in the race is not.
The descents
Climbing consumes the engine. Hard descending is often what destroys the legs.
- Col de Voza → Saint-Gervais: 8.28 km, −999 m, 38:53 (≈4:42/km)
- Grand Col Ferret → La Fouly: 9.7 km, −1,025 m, 46:52 (≈4:50/km)
- La Flégère → Chamonix: 6.56 km, −857 m, 33:48 (≈5:09/km)
The final descent was slower — about 10% per kilometre against the early Col de Voza descent. That is real fatigue, and there is no reason to hide it. What changes its meaning is the context: seventeen hours of racing, more than 160 km of load, repeated major climbs and descents, and a recent fall. Dhiman was also half an hour clear, while Olson was chasing Chassagne for second and missed him by 46 seconds. Their incentives on that hill were not the same.
So the point is not that he descended as fast at the end as at the start. He did not. It is that the deterioration stayed contained.
His cadence across the three went 176, 171, 167 spm. That does not prove preserved economy — turnover can hold while stride length, ground contact and force production all decay. It tells us only that turnover itself had not collapsed. That is enough, and there is no need to ask cadence to prove more than it can.
What about his heart rate?
The heart-rate data is considerably harder to interpret.
Segment averages range from 112 to 146 bpm. The highest came on more runnable ground — 146 from Saint-Gervais to Les Contamines, 143 from La Balme to Les Chapieux — while two of the steepest climbs read lower, at 119 to Bertone and 128 to Grand Col Ferret. The final descent reads 112.
It is tempting to call that extraordinary pacing control. I would not. Without knowing the sensor, its reliability through the event, and Dhiman’s own thresholds, heart rate cannot carry that argument. Cold, poles, wrist movement and changing skin contact all degrade wrist optical measurement — which is to say, the exact conditions of this race. The movement trace is much stronger evidence than the heart-rate trace.
What the data can and cannot support
The trace is useful. It is not laboratory data, and its own headline numbers do not fully reconcile.
The reported 169.38 km over 18:14:01 works out closer to 6:28/km than the 6:29 displayed. The 19 segment times do not sum exactly to the total. And the trace’s 169.38 km and +9,843 m sit against UTMB’s official 174 km and 9,700 m.
Its gradient figures are not elevation gain divided by distance either: 802 m over 5.28 km is about 15% net, while the platform reports 18.1%. That may reflect ascending-only portions, point-by-point sampling, or something else — the method is not published, so those percentages are reported gradients rather than survey measurements. The same caution applies to the distribution table above.
One recorded kilometre shows 352 m of ascent in 18:56. If both inputs are right, the resulting 1,116 m/h follows arithmetically; the uncertainty is whether GPS distance and elevation are precise enough on that terrain to trust either input.
The 18:46 it classifies as stopped time is similarly narrow. It means the algorithm rarely registered him as stationary — not that he spent only 18 minutes servicing himself, since a runner can walk an aid station, refill on the move or eat slowly without ever being counted as stopped.
None of this invalidates the dataset. It defines the question it can answer. The trace is good for comparing Dhiman with Dhiman — early against late, climb against descent, steep against runnable. It is not good for declaring any single metre, second or percentage exact.
What he says he built
Everything above is measurement. This section is Dhiman’s own account of his preparation, which is a different kind of evidence and worth labelling as such. It is interesting because it points at the same conclusion from the opposite direction.
By his description, speed was never the problem. He has talked about sitting comfortably around 3:30–3:40/km at roughly 125 bpm — an engine already built. The 2025 race gave him the diagnosis: second place in 19:51:37 after running with the leaders for a long time and then having a very hard finish. What he identified as the gap was not top-end pace but the ability to hold that level deep into the second half.
His stated answer was volume over intensity. Six-week blocks of around 25 hours, 200 km and more than 10,000 m of climbing per week. Over 100 hours of training in July, against under 60 the previous year. A winter base of some 305 hours and 170,000 m of vertical, much of it on skis, transferred gradually to running after Transvulcania. Around 30% more climbing than the same period in 2025 — and specifically more hiking, with poles and full kit, on steep technical ground in the Pyrenees, on the theory that walking is the efficient choice often enough to be worth training.
The racing followed the same logic. Rather than peaking for each event, he ran the 90K of Mont-Blanc on tired legs — second in 9:38:02 with roughly 180 km already in the week — because the point was to learn to perform fatigued rather than fresh. Then, about eighteen days out, Chamonix to Courmayeur: 75 km and 4,519 m in 8:18:14, arriving still feeling good.
None of that proves anything about the race itself; training plans are not results. But it is striking how precisely it matches what the segments show. He set out to fix the second half, and the second half is exactly where his 2026 race is remarkable.
Why this performance deserves admiration
Dhiman did not maintain one pace for eighteen hours. That was never possible. He went from 4:17/km to over 11 minutes per kilometre, from the 170s in cadence into the 120s and back, and his last major descent was slower than his first.
He fatigued.
What he retained was access to the movement the terrain required next. Run fast when the ground allowed it. Switch to a powerful hike when the grade demanded it. Reach the top still able to descend. Finish the descent still able to run. Repeat after 50 km, after 100, after 150. And when Chassagne cut into the lead between Champex-Lac and Trient, still have enough to open it again before Vallorcine.
No single mode of movement appears to have broken down catastrophically. In practical terms that is a very good picture of mountain-ultra durability: fatigue accumulated all day, but it never became a complete loss of whatever the course demanded next.
He fell with about 10 km to run, got up, finished the climb, and descended into Chamonix. Thirty-one minutes later, second place arrived.
The speed was already there in 2025. What he added was the capacity to keep it to the end — and that is what a finishing time compresses into one number. 18:16 is the result; the segments are the performance.
