Andreas Almgren won the New York City Marathon on November 1, 2026, in 2:06:01. It was his first marathon. He is a former data scientist, and he writes his own training plan. He has logged every session since 2011, he measures blood lactate in hard workouts, and he let those numbers decide how hard each day was allowed to be.
That is the reason this result matters to you. Almgren did not win New York on a secret workout. He won it with a feedback loop: a written plan, and a short list of numbers that decided when the plan could change. That loop used to need a coach and a lab. Today a heart rate strap plus an AI planning tool can run a version of it for an amateur. This feature explains how Almgren built his New York win, what the data shows, and how to run the same loop in a hybrid or HYROX week.
The training figures come from Almgrenโs COROS data and interviews as reported by Marathon Handbook on November 3, 2026, and from COROSโs own January 2026 profile. We have not seen his raw files.
Who he is, in numbers
| Fact | Figure |
|---|---|
| NYC Marathon 2026 | 1st, 2:06:01, 13 seconds ahead of Yemaneberhan Crippa. Halfway 1:04:04, second half 1:01:57 |
| Six weeks earlier | World half marathon champion in Copenhagen, 58:06, a European record |
| Education and work | Industrial engineering and management at KTH Stockholm. Worked as a data scientist at Svexa, a Swedish sports technology company |
| Training log | Every workout logged since 2011, kept in spreadsheets since 2018. A long-term plan that runs to 2029 |
| Coaching | Writes his own plan. Nike's Jordan Donnelly called him "99% self-coached". His longtime coach, Urban Aruhn, acts as a sounding board |
| Planning horizon | Picked New York after watching the 2025 race and started planning the block on the flight home, almost a year out |
| Marathon build | About 10 weeks from mid-August, 2,006 km, peak week about 241 km |
| Longest run before the build | About 32 km |
The loop behind the plan
Strip away the volume and Almgrenโs method is a loop with four steps. You can run the same loop at 40 km a week.
- Write the plan before the block starts. He fixes a structure and keeps it. โIโll keep a set structure for maybe six weeks, and adjust if needed,โ he told COROS. He also said, โonce you decide on a period, you should be quite consistent within that period.โ
- Measure a short list, every time. โI usually check four things: speed, lactate, heart rate, and overall feeling.โ That is the whole dashboard.
- Compare blocks, not days. He logs heart rate and speed for every session and compares them across weeks and training blocks. Repeating the same workouts week to week is what makes the comparison fair.
- Change one thing when the data says so. For the marathon he kept the double threshold structure and changed the parts the distance needed: longer long runs, heat sessions, fuel practice.
The point of the loop is that the number decides, not the mood of the day. Almgren knows from years of logs which lactate values he can repeat. That turns every hard session into a controlled test with a known ceiling.
| Signal | How Almgren uses it | Your version |
|---|---|---|
| Lactate | Finger-prick readings during threshold sessions. Caps around 1.5 to 2.5 mmol/L on double threshold days | A lab test once per block. Turn LT1 and LT2 into heart rate caps |
| Heart rate | Threshold sessions held at about 167 to 178 bpm. COROS measured his threshold heart rate at 168 in August | A chest strap on every hard session. Stay below your tested threshold heart rate |
| Speed | Splits on repeated sessions, for example 10 x 1 km at 2:34/km on 1 minute recovery | Run the same benchmark session every two to four weeks and compare the pace at the same heart rate |
| Feel | "I get a baseline for how I want workouts to feel" | Rate each session 1 to 10 in your log. A high rating at the usual pace is a warning |
What double threshold means
A double threshold day is two controlled threshold sessions on the same day, one in the morning and one in the afternoon. The point is volume at threshold without the damage of one long, hard session. Each session stays just under the point where lactate starts to climb fast, so you can recover by the next day.
For the marathon he made the morning session easier and the afternoon one longer:
- Morning: 50 minutes at the lower threshold (LT1). One treadmill run near 3:00/km read 1.36 mmol/L at a heart rate of about 140.
- Afternoon: longer intervals such as 16 x 1 km, or 5 x 2 km then 10 x 400 m. In the 16 x 1 km session on September 1 he ran the first eight at 2:50/km and read about 1.5 mmol/L, and the last eight at 2:47/km at about 2.5 mmol/L.
The lactate number is a ceiling. Eight weeks out from a Major, the world half marathon champion was holding himself down. He has said hard sessions should feel controlled and that he does not make all-out efforts in training.
His New York build, week by week
| When | Session or load | Why it matters for you |
|---|---|---|
| Mid-August | Long runs of 35 to 37 km, finishing with faster kilometres | Long runs end with work, not just distance |
| Week of August 31 | 233 km, two double threshold days, three 50-minute heat sessions at 38 ยฐC and 60% humidity, a 40 km progression run | Heat training sits on top of normal sessions, not instead of them |
| Week of September 7 | Peak, about 241 km. He had predicted 220 to 230 km in late August | The peak came early, eight weeks out |
| Flagstaff camp | Five-day cycle: hard long run, two easy days, double threshold day, easy day. Heaviest stretch 265 km in eight days | Two easy days after every hard long run |
| After Copenhagen | Fell ill, took three days fully off, week dropped to 87 km. Ran 35 km eight days after the race | Missing days is part of the plan, not a failure of it |
| Three weeks out | Last hard long run. Final long run 25 km instead of 30 km | A real taper, not a last-minute test |
| Race week | Arrived a week early, ran the last 6 km of the course several times. Practised about 120 g of carbohydrate an hour | Rehearse the finish and the fuel |
Three decisions the data made
The most useful part of Almgrenโs build is not a session. It is the calls he made when the plan and his body disagreed.
He held back on long runs. He ran long runs slightly easier than he could, so he recovered within two days. That let him fit in more of them. A long run you can repeat beats a heroic one that costs you a week.
He took three days off when he got sick. After Copenhagen he fell ill and stopped completely. His week fell to 87 km. Eight days after the race he ran 35 km and reported little loss of fitness. The log let him see that the block could absorb the gap.
He chose 25 km over 30 km for the last long run. About three weeks out he judged that arriving tired was a bigger risk than losing a little preparation. That is a decision about expected cost, made with a long history of how his body responds.
What the race data shows
COROS gave Marathon Handbook his race file. It shows a runner who stayed under his threshold for almost the whole race and spent his reserve at the end.
| Measure | First half | Second half | Final 600 m |
|---|---|---|---|
| Pace | About 3:02/km | About 2:56/km | 1:34, about 2:37/km |
| Average heart rate | 156 bpm, never above 171 | 163 bpm | Peak 178 bpm |
| Cadence | 178 steps/min | 183 steps/min | 195 steps/min |
| Ground contact time | 147 ms | 141 ms | 120 ms |
His heart rate stayed below his threshold of 168 for the first 37 km. It averaged 168 from 35 to 40 km and 174 from 40 km to the finish. Marathon Handbook puts his aerobic decoupling at about 1%, which means his heart rate barely drifted against pace. In Copenhagen he spent about 36 minutes above threshold. In New York it was about 14.
The race pace also matched the training data. His race paces correspond to training readings of roughly 1 to 1.5 mmol/L. He had run that intensity for weeks with a meter in his hand, so race day was a known quantity.
Where AI fits
Almgren is not reported to use an AI coach. He has a data science background and fifteen years of training logs. Most athletes have neither. This is where AI planning tools change the picture for amateurs: they can do the bookkeeping part of Almgrenโs loop for you.
A good AI training planner can:
- Build a block from your race date, your available days and your tested thresholds, not generic zones.
- Read your watch data after each session and flag when heart rate at a given pace drifts up across weeks.
- Rebuild the rest of the week when you miss a session or get sick, instead of leaving you to stack the missed work.
- Schedule test weeks so your zones come from a fresh lab test, an FTP test or a benchmark session.
What it cannot do is supply the inputs. An AI plan built on default zones and an incomplete log will be confidently wrong. Almgrenโs edge was the quality of his data and his discipline in following it. The tool only shortens the distance between the two.
Our editor, Thomas Prommer, runs this loop on himself. He has no human coach. He builds his sessions with aitrainingplan.app, anchored on thresholds from five lab lactate tests since 2024, and he tracks heart rate variability for readiness. In his own write-up of the Almgren method, his treadmill test put his aerobic threshold at 1.3 mmol/L and his anaerobic threshold at 2.8 mmol/L, the same band Almgrenโs 16 x 1 km session sat in. Disclosure: aitrainingplan.app is run by the same group that publishes this site.
How to scale it for a hybrid week
You are not running 240 km a week, and you have stations, sleds and a job. The pattern still transfers. These are our suggestions, not Almgrenโs:
- Get a real threshold first. Book a lactate test or run a field test, and set your zones from it. Every cap below depends on this number.
- Start with one double threshold day a week. Put it on a day with no heavy lifting. A simple version is 5 x 6 minutes in the morning and 8 x 3 minutes in the evening, both at an effort you could hold for about an hour.
- Use a cap, not a target. Without a lactate meter, cap the reps by heart rate. Stay below your lactate threshold heart rate. If the last rep is much harder than the first, you went too fast.
- Keep the rest of the week easy. Almgrenโs camp week had two easy days after every hard long run. In a HYROX block, your station and strength work counts as hard. Do not stack it on top of a threshold day.
- Fix the block for four to six weeks. Change one thing at a time and log the result. Speed, heart rate and how it felt are enough if you have no lactate meter.
- Repeat one benchmark session. Run the same session every two to four weeks and compare pace at the same heart rate. That is your version of his 10 x 1 km check.
- Practise race fuel at race effort. 120 g an hour is an elite figure. Asker Jeukendrupโs carbohydrate guidance suggests around 60 g an hour for two to three hours and up to about 90 g for longer efforts. Build up in training before race day.
What it means for HYROX
HYROX runs are 8 x 1 km between stations, run close to threshold effort for most athletes. A threshold day trains exactly that pace, and the second session of a double threshold day trains it on tired legs, which is what the last four runs feel like. Almgrenโs race file shows the other lesson: he stayed under threshold for 37 km and raced only the end. In HYROX that means holding your run heart rate under your threshold through the first five or six runs. If you want to see where your runs and stations land, try the HYROX finish predictor.
Sources and limits
Race: Marathon Handbook menโs results and Canadian Running. Training build, background, race file and lactate readings: Marathon Handbook, How Andreas Almgren trained for and won the New York City Marathon, November 3, 2026. Quotes on structure and data: COROS, Behind the training of Andreas Almgren, January 6, 2026. Copenhagen: World Athletics report. Self-coaching: Read The Profile, citing LetsRun. Reports disagree slightly on his peak week (about 235 to 241 km). Most Almgren statements in the Marathon Handbook piece are paraphrased, so we quote him directly only from COROS. No source reports that Almgren uses AI tools. The AI section and the scaling steps are our suggestions for amateur athletes, not his plan. This is not medical advice.
Want the race side of the story? Our partner site findmynextrace.com covers how the NYC debut was won and what it says about the course.
