6 Sleep Trackers, Ranked by Validation Evidence — Not Marketing

The yardstick every tracker is measured against

The reference standard for measuring sleep is polysomnography, the overnight test used in sleep clinics. Sensors on the scalp, face and body record brain waves, eye movements, muscle tone, breathing and heart rhythm, and a trained technician then labels the night in 30-second slices as wake, light sleep, deep sleep or REM sleep. A wearable has none of those brain signals. It works from movement and, in most cases, pulse, and its software estimates the rest.

Researchers compare the two using an agreement score called kappa, which runs from 0 (no better than chance) to 1 (perfect agreement). By convention, 0.21 to 0.40 is described as fair agreement, 0.41 to 0.60 as moderate, and 0.61 to 0.80 as substantial. It helps to know the ceiling: when two trained humans score the same night, they reach a kappa of about 0.78, a benchmark cited in an Australian validation of six wearables. Even the gold standard is not perfectly repeatable.

Which studies count, and for whom

Six lab studies carry most of the weight in this ranking, and they are easier to follow with short names. The Antwerp study (2025) tested six wrist devices in 62 adults, some with suspected sleep apnea. The Seoul study (2023) tested 11 trackers in 75 adults at two Korean centers. The Australian study (2022) put six devices on 53 healthy young adults at once. The Boston study (2024) compared a ring and two watches in 35 healthy adults. The Amherst study (2024) tested five devices in 53 healthy young adults, and the Berlin study (2025) tested rings in 45 sleep-clinic patients.

Funding shaped how each one was used. A study counts as independent for a brand only when neither its funding nor its authors' declared interests tie back to that brand's maker. The Boston study was funded by Oura, so it counts for Apple and Fitbit but not for Oura. The Australian team's research group receives funding and equipment from WHOOP, so that study counts for the other brands but not for WHOOP. The Seoul authors hold stock in the company behind a phone app that topped their own results; because that conflict concerns an app rather than any wearable, the study is counted here and flagged. Manufacturer white papers and product pages are used for prices, battery claims and software versions, and are always labeled as the company's own evidence.

How the 100 points are divided

Validation quality carries 40 points, and it measures how much is known rather than how good the results are. Up to 24 points come from the number of independent lab studies (three or more earns 24, two earns 16, one earns 8). Up to 16 come from how close the newest independently tested model is to the one on sale (one generation back earns 13, two earns 10, four earns 5, a different product line earns 2). No device earned the full 16, because no current model has been tested.

Sleep–wake accuracy carries 25 points, drawn only from independent studies. Twelve points reflect how far the device's total sleep time typically lands from the lab's (within 15 minutes earns 12, 16 to 30 minutes earns 8, more than 30 earns 3, and no published figure earns 6). Thirteen reflect how many of the lab's "awake" slices the device catches (half or more earns 13, 40 to 49 percent earns 9, 30 to 39 percent earns 5, less than 30 earns 2). Where several studies exist, the middle value is used.

The remaining 35 points cover ownership. Three-year cost (hardware plus any subscription the device needs) carries 10: a total of $400 or less earns 10, $401 to $550 earns 7, and anything higher earns 4. Privacy and data access carries 10: four for end-to-end encryption, up to three for publishing a transparency report and promising to notify users of government data requests, and three for a documented way to export your data, with the first two drawn from a July 2026 survey of wearable makers by the Electronic Frontier Foundation. Battery life carries 10, using the upper end of each maker's claim (a week or more earns 10, three to six days earns 7, 30 to 48 hours earns 4, and 24 hours or less earns 2). Comfort carries 5 (a ring earns 5, a screenless strap 4, a watch 3, and a watch over 40 grams 2), and it is only a rough stand-in based on shape and weight, since we did not wear-test these devices.

What is left out of the score on purpose

Sleep-stage estimates appear in the comparison table but earn no points. Across the independent studies, stage agreement for these brands ranged from a kappa of 0.20 to 0.60, and the same device often scored very differently from one lab to the next. When the gap between studies is wider than the gap between devices, a ranking built on those numbers would mostly reflect which study a device happened to land in.

Ties are broken by validation quality first, then by sleep–wake accuracy. Prices are US list prices before sales tax, which varies by state and applies to all six.

Here is the full picture in one place, in countdown order. Sleep figures describe the earlier-generation devices that were actually tested, taken from independent studies only. Price, battery and subscription details describe the current model.

One pattern runs through every row, and it is the most useful thing to carry into the rest of this piece. These devices are very good at recognizing sleep and much less good at recognizing wakefulness. In the Antwerp study of six wrist-worn trackers, every device correctly labeled more than 90 percent of the lab's sleep slices, yet they caught only 29 to 52 percent of the slices when the person was actually awake.

That lopsidedness has a practical consequence. Lying still and awake looks a lot like sleep to a sensor on your wrist, so the error runs in one direction: trackers tend to credit you with more sleep than you got. The effect is smallest for people who fall asleep quickly and stay asleep, and largest for people who spend long stretches awake in bed, which is often the group most interested in tracking.

What the evidence shows

Samsung's one independent test went well. In the Seoul study of 11 consumer sleep trackers, the Galaxy Watch 5 reached a kappa of 0.42 for sleep stages, level with the best wearable in that group. It caught 48 percent of wake slices, the highest of the five wearables tested, and its sleep efficiency figure differed from the lab's by just 0.4 percentage points.

Where the gaps are

The difficulty is how little stands behind those numbers. Only 22 people wore the Samsung watch in that study, all of them in Korea, and the group as a whole had a fairly high rate of disordered breathing during sleep. We also did not find a per-device total sleep time difference in the published results, which is why that part of the score sits at the midpoint.

The only other lab test we found does not count as independent. A 2023 validation of the Galaxy Watch 3 in 32 healthy adults reported that the watch detected 95.4 percent of sleep and 52.4 percent of wake, but it was supported by a Samsung Medical Center grant and used watches supplied by Samsung. The Watch9 itself sits four generations past the newest model anyone has tested independently, and six past that Samsung-supported one.

Trade-offs beyond accuracy

The Watch9 has the lowest entry price of the six and no subscription for sleep tracking. According to Samsung's US product page, it starts at $379.99, claims up to 40 hours of battery life, and needs a compatible Android phone to set up, with some features reserved for Samsung Galaxy phones. It does not work with an iPhone.

Privacy is where this entry is least settled. Samsung was not among the ten companies in the Electronic Frontier Foundation's survey, so it earns points only for its documented data download, and its low privacy score reflects missing information rather than a known problem. The watch also offers a sleep apnea feature that Samsung describes as regulated by the FDA as a medical device and intended to notify undiagnosed users of risk. That is a separate function from sleep staging, it is not a diagnosis, and it sits outside this ranking.

Score: 48 out of 100. Validation quality 13 of 40, sleep–wake accuracy 15 of 25, three-year cost 10 of 10, privacy and data access 3 of 10, battery 4 of 10, comfort 3 of 5.

What the evidence shows

Garmin has something Samsung lacks, which is repetition. Three independent labs tested two different Garmin models, and they landed in nearly the same place. In the Amherst study of five commercial trackers, the Vivosmart 4 overestimated total sleep time by 41.7 minutes and was the only device that did not match research-grade equipment on that measure. The Antwerp study found 38.4 minutes of overestimate on the same model, and the Australian study found 43.8 minutes on a Forerunner 245.

The reason shows up in the wake numbers. Garmin's devices caught 29 percent of wake slices in Antwerp and 27 percent in Australia, the lowest or close to it in each study. On an average night in Antwerp, the lab recorded 42 minutes awake after falling asleep, and the Garmin reported four. Stage agreement was fair, with kappas of 0.21 and 0.25.

Where the gaps are

Garmin's own evidence tells a more flattering story, and the difference is instructive. A 2019 sleep study published on Garmin's blog reported a kappa of 0.54 for an earlier Vivosmart, but it was run by Garmin, was not peer reviewed, and compared the watch with a portable brain-wave device rather than full polysomnography.

There is also a fair point in Garmin's favor. The Venu 4 belongs to a different and much newer product line than the two models that were tested, and its software has had years to change. It may well do better. There is simply no independent study yet to show it, which is why it earns the fewest points for closeness.

Trade-offs beyond accuracy

On ownership, the Venu 4 is strong. Garmin lists it at $549.99 with up to 10 days of battery life on the 41 mm model, it works with both iPhone and Android, and sleep score and sleep stages need no subscription. The optional Connect+ plan, at $6.99 a month, is not required for any of it. At about 45 grams, it is the heaviest device here.

Privacy pulls the score down. The EFF survey found no end-to-end encryption, no transparency report and no public promise to notify users of government requests. Garmin does provide a full data export through its account center.

If your interest is as much in training as in sleep, our guide to fitness trackers matched to different goals covers that side of the decision.

Score: 53 out of 100. Validation quality 26 of 40, sleep–wake accuracy 5 of 25, three-year cost 7 of 10, privacy and data access 3 of 10, battery 10 of 10, comfort 2 of 5.

What the evidence shows

WHOOP's newest tested model is only one generation behind the one on sale, the smallest gap of any device here. In the Antwerp study, 40 people wore a WHOOP 4.0. It overestimated total sleep time by 24.5 minutes on average, caught 40 percent of wake slices, and reached a kappa of 0.37 for stages. It did stand out in one respect: it correctly labeled 70 percent of the lab's deep sleep slices, the best of the group, although it also overestimated the nightly deep sleep total by about half an hour.

A second number in that study deserves a moment. WHOOP's typical per-night error for total sleep time was 50 minutes. An average that is 24 minutes off can hide individual nights that miss by much more, in either direction.

Where the gaps are

WHOOP is credited with just one independent study, and the reason is a conflict of interest that is easy to miss. The Australian study looked favorable for the WHOOP 3.0, with total sleep time within about 12 minutes of the lab's. But as the authors' own disclosure in that paper explains, their research group receives funding and equipment from WHOOP, and WHOOP was one of only two companies that supplied its data directly to the researchers. The study itself was funded by the Australian Institute of Sport, and the authors state that WHOOP had no role in it. Under the rule used here, it still does not count toward WHOOP's independent total.

Trade-offs beyond accuracy

WHOOP is priced differently from everything else on this list. According to WHOOP's membership page, the device is included with a membership that starts at $199 a year, and it does not function without one. Three years on that tier comes to $597, which is almost exactly what an Oura Ring 5 costs with three years of its own membership. The difference is what remains if you stop paying. With WHOOP, the answer is nothing.

In return, the strap has no screen to light up at night, a stated battery life of more than 14 days, and CSV exports of sleep data that open in any spreadsheet. On privacy, WHOOP promises to notify users of law-enforcement requests but offers neither end-to-end encryption nor a transparency report.

Much of what WHOOP reports each morning leans on overnight heart rate, and our explainer on what resting heart rate can tell you gives useful background on that signal.

Score: 61 out of 100. Validation quality 21 of 40, sleep–wake accuracy 17 of 25, three-year cost 4 of 10, privacy and data access 5 of 10, battery 10 of 10, comfort 4 of 5.

What the evidence shows

Apple's results depend heavily on which lab you ask. In the Antwerp study, the Series 8 posted the best stage agreement of all six wrist devices (kappa 0.53), the best wake detection (52 percent) and the smallest typical per-night error for total sleep time, at just under 28 minutes. In the Seoul study, the same model had the lowest stage agreement of the five wearables (kappa 0.30).

One finding is consistent across labs, and that is how Apple handles deep sleep. In the Boston comparison of three wearables, the Series 8 reported an average of 51 minutes of deep sleep on nights when the lab measured 94, and it labeled the difference as light sleep instead. Antwerp found the same direction of error. Anyone comparing deep sleep totals between an Apple Watch and another brand is likely to see a gap that comes from the software rather than from their sleep.

It is also worth knowing that the Apple Watch produced usable recordings for only 20 of the 35 people assigned to it in Antwerp, and for 29 of 35 in Boston.

Where the gaps are

Apple is unusually open about how its sleep tracking works. Apple's sleep-stage white paper explains that the classifier is built on the watch's motion sensor, reports a kappa of 0.63 in Apple's own validation group of 166 people, and says a 2025 software update raised that to 0.68. It also states plainly that the stages are not intended for clinical use.

That update is the catch. Every independent study tested the earlier software, and Apple measured the new version on its own existing dataset. So the Series 12 carries two gaps at once: hardware four generations past the tested model, and a sleep algorithm that no outside lab has examined. The outside evidence is thin in general. A 2026 review of Apple Watch accuracy research found that only three of its 82 included studies had tested sleep-stage classification.

Trade-offs beyond accuracy

Privacy is where the Apple Watch separates from the field. In the EFF's survey of wearable makers, sleep data stored in Apple's Health app was the only wearable data protected by end-to-end encryption, switched on by default, and Apple was one of two companies publishing a transparency report. It is the only device here to earn all 10 privacy points.

The weak spot is battery. Apple lists the Series 12 at $399 with up to 24 hours of everyday use, or 38 in Low Power Mode, which means finding a daily charging window that is not the night. Apple says 15 minutes on the charger adds up to 12 hours. The watch needs an iPhone, requires no subscription, and exports everything in the Health app as a single XML file.

Apple's sleep features lean on duration and timing, and if that is the question on your mind, we have looked separately at whether waking early or sleeping longer serves you better.

Score: 71 out of 100. Validation quality 29 of 40, sleep–wake accuracy 17 of 25, three-year cost 10 of 10, privacy and data access 10 of 10, battery 2 of 10, comfort 3 of 5.

What the evidence shows

The Pixel Watch's sleep tracking grew out of Fitbit's, and Fitbit-branded devices are the most frequently tested family in this research. The picture is steady. Total sleep time landed within 6 to 11 minutes of the lab's in the Antwerp study, a difference small enough that it was not statistically distinguishable from zero. It was within 3 minutes in Boston and within 9 to 11 minutes in Amherst. Stage agreement was moderate across the board, with kappas between 0.40 and 0.55.

Four studies in three countries, across five different Fitbit models, pointing the same way is about as much consistency as this field offers. It is the main reason this entry ranks second.

Where the gaps are

The caveat is that almost all of that evidence comes from Fitbit-branded hardware. Treating those results as a guide to the Pixel Watch is an assumption, and the one study that tested a Pixel Watch directly gives reason for care. In the Seoul study, the first-generation Pixel Watch caught only 23 percent of wake slices and overstated sleep efficiency by 12.8 percentage points, the largest overestimate among all 11 trackers tested. Its stage agreement, at a kappa of 0.40, was in line with the Fitbit devices.

The Pixel Watch 5 is four generations on from that device. In Google's launch announcement, the company says the new watch delivers "15% more accurate sleep stage detection." The post does not say what that is measured against or point to a study, so for now it is a manufacturer's claim rather than evidence.

Trade-offs beyond accuracy

Google prices the Pixel Watch 5 at $399 for the 41 mm size and $429 for the 45 mm, with stated battery life of up to 30 and 40 hours. It weighs about 31 grams in the smaller size and works with phones running Android 12 or newer, not with iPhones. As of early October 2026, sleep score and sleep stages are part of the free tier, and the optional Google Health Premium plan costs $9.99 a month.

On privacy, Google publishes a transparency report and promises to notify users of government requests, though it does not offer end-to-end encryption for this data. An export is available through account settings.

A tracker can show you a pattern, and what to do with one is a different subject, covered step by step in our evidence-based sleep hygiene framework.

Score: 73 out of 100. Validation quality 29 of 40, sleep–wake accuracy 21 of 25, three-year cost 10 of 10, privacy and data access 6 of 10, battery 4 of 10, comfort 3 of 5.

What the evidence shows

Oura comes first because more is known about it than about any other device here, and because its tested model is only two generations old. Four studies with no ties to the company have examined an Oura ring. Their most dependable finding concerns sleep duration. Average total sleep time was within 2 minutes of the lab's in the Australian study and within 14 in Amherst. In the Berlin study of sleep-clinic patients, it was within 12 minutes even in people with diagnosed sleep disorders, a much harder group to measure.

Wake detection in the independent studies ranged from 38 to 57 percent, similar to the better watches.

Where the gaps are

Oura's best-known numbers do not come from those studies. The Boston comparison, where the ring reached a kappa of 0.65 and outperformed both watches, was paid for by the company. As the funding and conflict statements in that paper disclose, the research was funded by Oura Ring Inc. and the lead author sits on Oura's medical advisory board. A larger Japanese study of 96 people that reported strong results was also mainly funded by Oura. Neither is counted in the accuracy score.

The independent picture for sleep stages is more modest. Stage agreement was a kappa of 0.35 in Seoul and 0.31 in Berlin, and in both, the ring assigned the correct stage a little over half the time. The Berlin authors concluded that the output was not sufficient for individual diagnostic use.

That makes one marketing line worth a close read. Oura's Ring 5 product page advertises 95 percent sleep staging accuracy compared with a clinical sleep lab. The footnote attached to that claim says the figure comes from two-stage classification, which means telling sleep from wake. It is not a claim about telling light, deep and REM sleep apart, and the page also notes the ring is not intended for use as a medical device.

Trade-offs beyond accuracy

The Ring 5 starts at $399, weighs from 2 grams, lists battery life of up to 6 to 9 days, and works with both iPhone and Android. It is the least noticeable device here to sleep in, by shape alone.

Cost is the real trade-off. Oura's membership is $5.99 a month or $69.99 a year, and without it the app shows only three daily scores. Over three years, the ring and membership together come to about $609, the highest total on this list. On privacy, Oura began promising to notify users of law-enforcement requests in June 2026, but it does not offer end-to-end encryption or publish a transparency report.

So the first-place finish needs a plain explanation. Oura leads on how much independent evidence exists, how recent it is, and how close its sleep duration figures have been. It does not lead because its deep sleep and REM numbers are proven to be better than anyone else's.

Score: 79 out of 100. Validation quality 34 of 40, sleep–wake accuracy 21 of 25, three-year cost 4 of 10, privacy and data access 5 of 10, battery 10 of 10, comfort 5 of 5.

What it can tell you

It can tell you which brands have opened themselves to outside testing, who paid for the studies behind the headline claims, and how many product generations have passed since. It can also tell you which numbers on the screen have earned the most trust. Across every study, total time asleep was the most dependable figure, time awake during the night was consistently undercounted, and deep sleep and REM were the least certain of all.

Different goals point to different rows of the table. Someone who mainly wants a steady record of sleep duration has several reasonable options near the top. Someone for whom data privacy comes first may weigh the Apple Watch's encryption more heavily than its battery. Someone watching the three-year total may look first at the devices with no required subscription, and someone who dislikes charging may start with the three that go several days or more between charges.

What it cannot tell you

It cannot tell you how the specific model in your shopping cart performs, because no one outside these companies has published that test. It cannot tell you how accurate any single night will be, since devices that look close on average were often off by half an hour or more on individual nights. And the results come mostly from healthy adults sleeping one night in a lab, which is not how most people sleep at home.

Most importantly, none of these devices can tell you whether you have a sleep disorder. The American Academy of Sleep Medicine's position, first issued in 2018, is that consumer sleep devices are unable to diagnose sleep disorders, and that ongoing sleep problems or daytime sleepiness are worth raising with a licensed medical provider regardless of what a wearable reports.

That does not make tracking pointless. In a 2026 position statement, the National Sleep Foundation took the view that evidence-based consumer sleep technology may improve sleep health when it is used appropriately, and named understanding its limits as the first step.

There is one more thing worth knowing before you buy, which is that the score itself can become a source of worry for some people. We explored that in why a smartwatch sleep score can make sleep worse.

A sleep tracker at its best is a patient, slightly nearsighted observer. It sees the broad shape of your nights clearly, misses some of the detail, and gets more useful the longer you watch the trend rather than the single reading. Knowing where each one has been checked, and where it has not, puts you in a good position to choose the one that fits your life.