The short answer: at the three big autumn majors of 2026, only 9.0% of finishers ran a negative split, and the typical runner's second half was 12 minutes slower than the first. Even among sub-3 finishers, fewer than one in five managed a faster second half. The slowdown is not spread evenly: almost everyone holds pace to 20 km, and 61% of runners had their slowest 5 km between 35 and 40 km. Between the 5-15 km stretch and the 30-40 km stretch, 65% of the field slowed by more than 30 seconds per kilometre and 43% by more than a minute per kilometre.
Those numbers come from the official timing-mat splits of 163,733 finishers at the 2026 Berlin, Chicago and New York City marathons, not from a survey or a sample of watch uploads. This is the same population as the marathon finish time distribution guide, so the two sets of numbers line up.
How many runners negative split a marathon?
Across the whole field, 9.0% of runners finished the second half faster than the first. Another 12.0% ran the two halves within two minutes of each other, which is about as even as a marathon gets. Everyone else, roughly four runners in five, slowed noticeably.
The share of negative splits falls steadily with finish time:
| Finish time | Finishers | Negative split | Halves within 2 min | Median second half slower by | Second half 10+ min slower |
|---|---|---|---|---|---|
| Under 3:00 | 8,775 | 18.6% | 43.0% | 2.3 min | 6.7% |
| 3:00-3:30 | 19,475 | 14.7% | 25.0% | 4.6 min | 21.8% |
| 3:30-4:00 | 32,481 | 14.1% | 18.0% | 6.6 min | 34.8% |
| 4:00-4:30 | 30,419 | 8.5% | 8.8% | 12.0 min | 57.1% |
| 4:30-5:00 | 27,936 | 6.6% | 5.6% | 16.7 min | 69.9% |
| 5:00-5:30 | 18,516 | 3.6% | 2.7% | 22.7 min | 82.9% |
| 5:30-6:00 | 13,292 | 2.7% | 1.7% | 26.7 min | 87.9% |
| Over 6:00 | 12,839 | 1.6% | 1.3% | 30.3 min | 92.1% |
Two things stand out. First, the sub-3 group is the only one where a large minority (43%) ran close to even. Second, the break point sits right around four hours: a 3:30-4:00 finisher typically loses 6.6 minutes in the second half, a 4:00-4:30 finisher loses almost twice that. Among runners over five hours, the second half is routinely 20 to 30 minutes slower than the first.
Women were slightly more likely to negative split than men (9.9% against 8.4%), and the flat Berlin course produced more negative splits (11.0%) than Chicago (7.9%) or hilly New York (8.4%).
The fade curve: where each group loses time
Every group ran the 5-15 km stretch slightly faster than the first 5 km, by about 1%, which is the crowd thinning out and runners settling in. From 15 km the lines split apart.
Sub-3 finishers barely move until 30 km and are only 5% slower than their opening pace at 35-40 km. The 3:30-4:00 group holds to 20 km, drifts 2% slower by 30 km, then slows 10% over 35-40 km. The 4:00-4:30 group is already 3% slower at 25 km and ends up 16% slower over 35-40 km. Runners finishing over five hours are 15% slower by 25 km and 26% slower by 40 km, which means a runner who opened at 6:30/km is walking parts of the last 10 km at over 8:00/km.
For runners finishing within five minutes of 4:00 (11,915 of them), the median segment was 2% slower than the opener at 25 km, 3% at 30 km, 6% at 35 km and 10% at 40 km. If your goal is sub-4 and your first 5 km is on 5:35/km, the data says expect to be running around 6:10/km through 35-40 km unless you do something differently from most people.
Nearly every group speeds up a little from 40 km to the finish. That is the finishing kick, and it is real even for the slowest runners, but it recovers only a fraction of what was lost.
The wall, in seconds per kilometre
Comparing a runner's pace over 30-40 km with their pace over 5-15 km gives a single number for the wall. The median finisher slowed by 50 seconds per kilometre. Only 6.9% of runners held or improved their early pace through 30-40 km.
| Finish time | Median pace 5-15 km | Median pace 30-40 km | Median slowdown | Slowed by 30+ s/km | Slowed by 60+ s/km |
|---|---|---|---|---|---|
| Under 3:00 | 3:56/km | 4:10/km | 10 s/km | 13.4% | 1.9% |
| 3:00-3:30 | 4:31/km | 4:55/km | 20 s/km | 34.3% | 10.1% |
| 3:30-4:00 | 5:08/km | 5:41/km | 29 s/km | 48.2% | 20.6% |
| 4:00-4:30 | 5:40/km | 6:33/km | 50 s/km | 69.4% | 41.5% |
| 4:30-5:00 | 6:16/km | 7:25/km | 68 s/km | 78.5% | 55.7% |
| Over 5:00 | 7:16/km | 8:59/km | 97 s/km | 90.0% | 75.2% |
The chart shows two very different shapes. Finishers under 4:00 cluster tightly around a 10 to 20 s/km slowdown. Finishers over 4:00 are spread across the whole range, with a peak near 55 s/km and a long tail: three quarters of the over-5:00 group slowed by more than a minute per kilometre.
Where it happens is consistent. The slowest 5 km segment of the race was 35-40 km for 60.8% of runners and 30-35 km for another 18.7%. Fewer than 10% of runners had their slowest segment before halfway.
What this means for your pacing plan
The data is not an argument that negative splits are impossible. It is evidence that almost nobody achieves one by accident, and that the typical runner opens faster than they can hold. A few practical conclusions:
- Your first 5 km is a trap. Every group ran kilometres 5 to 15 faster than kilometres 0 to 5, then faded. Treat the opening pace as a ceiling, not a target.
- Plan for the fade you can afford, not zero fade. A 4:00 runner who plans even 5:41/km splits and then slows 10% from 35 km will finish around 4:04. Opening at 5:45/km and holding it leaves more in the tank for the last 7 km.
- Even-ish is the realistic goal. Of the 8,775 sub-3 finishers, 43% ran their halves within two minutes of each other. That, not a dramatic negative split, is what good pacing looks like at the sharp end.
- Build the plan around 30-40 km. That is where 80% of runners hit their slowest patch. Fuel earlier than feels necessary and save any surge for after 35 km.
The pace calculator lets you type a plan like "marathon in 3:55, slight negative split" and see the per-km splits, or "make the last 7 km 15 seconds slower" to model a realistic fade.
Where the data comes from
The splits are the official 5 km timing-mat results of the 2026 BMW Berlin Marathon, Bank of America Chicago Marathon and TCS New York City Marathon, taken from the open Pace Notes dataset (CC BY 4.0), which compiles verified results from race organisers. Of the 165,113 men and women who finished the three races, 164,067 had all eight 5 km splits recorded; 334 were dropped for an implausible segment (a 5 km split faster than 12:00 or slower than 75:00, almost always a missed mat), leaving 163,733.
Halfway time is estimated as the 20 km mat time plus 1.0975 km at the runner's 20-25 km pace, since the dataset's half-marathon split field is empty for these races. The "wall" measure compares average pace over 30-40 km with average pace over 5-15 km, which ignores the first 5 km (crowded starts) so that congestion does not show up as a fade.
Frequently asked questions
What percentage of marathon runners negative split?
9.0% of finishers at the 2026 Berlin, Chicago and New York City marathons ran a faster second half. The share ranges from 18.6% of sub-3 finishers to 1.6% of finishers over 6 hours.
How much slower is the second half of a marathon for most runners?
The median second half was 12.2 minutes slower than the first across all finishers. For 3:30-4:00 finishers it was 6.6 minutes, for 4:00-4:30 finishers 12.0 minutes, and for runners over 5:00 it was 20 to 30 minutes.
Where do marathon runners hit the wall?
Between 30 and 40 km. 60.8% of runners had their slowest 5 km from 35 to 40 km and another 18.7% from 30 to 35 km. Fewer than 10% had their slowest segment in the first half.
Is a positive split always bad?
No. A small positive split is the norm even at the front: sub-3 finishers typically ran their second half 2.3 minutes slower. The problem is the size of the fade. Slowing by 10 to 20 seconds per km over the last 10 km costs a few minutes; slowing by a minute per km, which 43% of the field did, costs ten or more.
Does a flat course help with pacing?
A little. Berlin had the highest negative-split share of the three races (11.0%) and the smallest median second-half slowdown (10.1 minutes), against 14.0 minutes at Chicago and 12.4 minutes at New York. Course profile matters less than opening pace.