The most accurate sleep trackers, against polysomnography

Almost nothing here has been measured

Your tracker shows you a confident four-color hypnogram with minute counts for light, deep and REM. The number of consumer devices for which that graph has been checked against clinical polysomnography in a published, peer-reviewed head-to-head is three.

Robbins et al., Sensors 24(20):6532 (2024) — from Brigham and Women's Hospital — put 35 healthy adults through a single-night inpatient study wearing an Oura Ring Gen3, a Fitbit Sense 2 and an Apple Watch Series 8, all scored against full polysomnography at 30-second epochs per AASM guidelines.

Sleep versus wake: everything works

Oura 92.0% agreement (κ = 0.60), Fitbit 91.0% (κ = 0.52), Apple 93.0% (κ = 0.60). If all you want is total sleep time, any modern tracker is adequate and you can stop reading and keep your money.

Four-stage classification: differences appear

Oura 76.3% (κ = 0.65), Apple 75.0% (κ = 0.60), Fitbit 70.9% (κ = 0.55).

Deep sleep: the numbers that should change your mind

Headline agreement hides the interesting part. Deep sleep is the stage people actually care about and modify their behavior over, and sensitivity to it varied enormously:

• Oura Ring: 79.5%
• Fitbit Sense 2: 61.7%
• Apple Watch Series 8: 50.5%

And in nightly summary terms: the Apple Watch overestimated light sleep by 45 minutes and underestimated deep sleep by 43 minutes per night. Fitbit overestimated light by 18 and underestimated deep by 15. Oura showed no significant difference from polysomnography on seven of eight measures, missing only sleep latency by five minutes.

If you have been anxious about your deep sleep numbers on an Apple Watch, that result is probably the most useful thing on this page. The device is measuring roughly half of it.

The limits of that study, stated properly

A study quoted past its evidence is marketing with a citation attached, so here are the authors' own caveats:

• One night, in a laboratory, not at home.
• 35 healthy adults aged 20-50, without insomnia. Not a clinical population.
• Devices were compared only during scheduled sleep episodes rather than across 24 hours, which the authors note may inflate agreement.
• Device failures: Apple Watch data was unusable for 6 of 35 participants, Fitbit for 2.
• It tested those three devices in those generations, and the authors say explicitly the findings do not generalize to other devices or later models.

That last point matters here more than anywhere: the Oura result belongs to the Gen3, not the Ring 4 and certainly not the Ring 5.

The 76.3% versus 79% discrepancy

Worth a paragraph of its own. Oura's press release about this study described it as finding 79% four-stage agreement. The paper reports 76.3%.

It is a small gap and it does not change the conclusion — Oura still came first. But it is a precise illustration of why we read primary sources and tell you which one we read. Every number on this page comes from the paper.

Why the Ring 5 ranks below the Ring 4

This is the rubric doing something that looks strange and is, we think, correct.

The Oura Ring 5 is newer, smaller, and probably better. It ranks below the Ring 4 here because the peer-reviewed result belongs to the Gen3, and a new sensor running a new algorithm is precisely what a validation study exists to test. Assuming the result carries forward is assuming the thing the study was built to check.

If Oura publishes a Ring 5 validation, its score moves. That is how this is supposed to work.

Garmin's absence, and what it does and does not mean

Garmin is one of the largest wearable makers in the world and was not in this comparison at all. We could find no published PSG head-to-head of Garmin's sleep staging.

That is not evidence that Garmin is inaccurate. It is an absence of evidence — and under our hierarchy an absence caps how high a device can rank on accuracy grounds, which is why the Venu 3 and the Index Sleep Monitor sit mid-table despite being good products. We would rather say that plainly than rank them on reputation.

What we could not read

Two relevant papers were behind publisher blocks that returned errors to our fetch on 24 July 2026: the Svensson multi-night Oura validation in Sleep Medicine, and a 2025 systematic review and meta-analysis in OTO Open. A pooled meta-analysis would be stronger evidence than the single study we rely on, so this is a real gap rather than a formality. We do not quote numbers from papers we have not read, so neither influences any score on this page. Both are listed in the sources below, marked unread.