VO2 max accuracy, device by device

Why this table is mostly Apple

Not because Apple is worse. Because Apple gets measured. Researchers pick the device that is easiest to recruit participants for, and in this category that is the Apple Watch — so the brand with the most published data is also the brand whose flaws are documented, while its competitors sit in a silence that reads, unfairly, as competence.

This site treats that silence as its own category rather than a pass. Under our evidence hierarchy, a device with an unflattering measurement can still outrank a device with nothing at all — which is why the Apple Watch Ultra 3 scores above several better-specified watches on the evidence column.

The pattern in the numbers

Both Apple validations point the same way: the estimate is systematically low, and the fitter you are, the lower it goes. In the 2024 study the least fit group saw the watch overshoot by nearly 4 mL/kg/min, while the fittest group saw it undershoot by 12 — the rough difference between a competent club runner and an elite one.

The mechanism is not mysterious. A watch infers aerobic capacity from the relationship between pace and heart rate, and at the top end that relationship stops discriminating: highly trained runners differ from each other in running economy and lactate handling, neither of which pace-and-heart-rate can see. That reading is ours, not a finding either paper states.

The threshold rows, and why they are here

Lactate threshold is not VO2 max, but it is the closest measured neighbor and it is estimated by the same class of model from the same inputs. The 2025 threshold study is the only peer-reviewed measurement of a current Garmin or COROS training estimate that we have found, and both watches overestimated pace by roughly a quarter.

One caveat the authors publish and we repeat: the device samples were unequal — 100 runners wore the Huawei, 23 the Garmin, 17 the COROS — and the Garmin protocol required an ECG chest strap while the other two used wrist sensing. Full detail is on our lactate threshold page.

What would actually count as evidence

A validation of a VO2 max estimate needs three things, and press releases usually have none of them: a laboratory reference (indirect calorimetry or a gas analyzer), a participant sample that spans fitness levels rather than clustering at one end, and published limits of agreement rather than a correlation coefficient. Correlation says two numbers move together; limits of agreement say how far apart they are on any given day, and that is what you need to know before quoting your VO2 max to anyone.

When a manufacturer publishes that, we will put it in this table with its name on it.

How to use the number you have

• As a trend against itself. A consistent bias still tracks real change. If your estimate rises 3 points over a training block on the same watch, something real probably happened.
• Never across brands. Two watches on the same wrist will disagree, and the disagreement tells you about the models, not about you.
• Never as a figure to quote. Both Apple studies say plainly that the estimate is not adequate for decisions of that kind. If the actual value matters, get tested.