Sleep Tracker Accuracy Leaderboard

Q: What is the most accurate sleep tracker?

In Miller et al. (2022), Somfit had the highest sleep-stage agreement with polysomnography (65%, κ 0.52), followed by Oura (61%) and WHOOP (60%). Among wrist and ring wearables, Oura and WHOOP led.

Q: Can wearables track sleep stages accurately?

Only moderately. The authors concluded all six devices were valid for sleep timing and duration but require improvement for specific sleep stages.

Reviewed by Sameer Sontakey, Co-founder and Chief AI & Algorithms Officer at Sensor Bio. ChatPPG is independent and device-neutral; it is not a Sensor Bio product.

In a 53-adult polysomnography study (Miller et al., 2022), Somfit (65%, κ 0.52), Oura (61%, κ 0.43) and WHOOP (60%, κ 0.44) had the highest sleep-stage agreement. All six were valid for sleep timing, but none was strong at staging.

Which wearable is most accurate for sleep tracking?

84 published results from 15 studies in the wearable accuracy database. Each ranking below compares devices only on the same statistic against the same kind of reference. Results from different statistics or references are never ranked together.

Evidence by brand

Apple Watch4 studies, n = 176
Oura10 studies, n = 501
WHOOP4 studies, n = 133
Garmin4 studies, n = 198
Fitbit5 studies, n = 176
Pixel Watch1 study, n = 30
Samsung2 studies, n = 67
Polar2 studies, n = 102
Withings1 study, n = 62
AmazfitNo published validation found
UltrahumanNo published validation found
COROSNo published validation found

Sleep stages (4-stage): epoch-by-epoch agreement (%)

Reference: Polysomnography (PSG). Higher is better. 5 studies; a device tested in more than one study shows its median.

1Somfit65%1 study, n = 53 (65%) · Miller et al. (2022)
2Samsung Galaxy Watch564.9%1 study, n = 22 (64.9%) · Lee et al. (2023)
3Fitbit Sense 264.6%1 study, n = 26 (64.6%) · Lee et al. (2023)
4WHOOP (strap, 2.0 era)64%1 study, n = 12 (64% (κ 0.47)) · Miller et al. (2020)
5Google Pixel Watch63.5%1 study, n = 30 (63.5%) · Lee et al. (2023)
6WHOOP (auto-detect)63%1 study, n = 6 (63%) · Miller et al. (2021)Maker funded
7Oura Ring Gen261%1 study, n = 53 (61%) · Miller et al. (2022)
8WHOOP 3.060%1 study, n = 53 (60%) · Miller et al. (2022)Maker funded
9Apple Watch Series 856.4%1 study, n = 26 (56.4%) · Lee et al. (2023)
10Oura Ring Gen354.3%1 study, n = 53 (54.3%) · Lee et al. (2023)
11Oura Ring53.2%1 study (53.18%) · Herberger et al. (2025)
12Apple Watch Series 653%1 study, n = 53 (53%) · Miller et al. (2022)
13Polar Vantage V51%1 study, n = 53 (51%) · Miller et al. (2022)
14Garmin Forerunner 245 Music50%1 study, n = 53 (50%) · Miller et al. (2022)

Sleep stages (4-stage): cohen's kappa

Reference: Polysomnography (PSG). Higher is better. 4 studies; a device tested in more than one study shows its median.

1Somfit0.521 study, n = 53 (κ = 0.52) · Miller et al. (2022)
2WHOOP (strap, 2.0 era)0.471 study, n = 12 (κ = 0.47) · Miller et al. (2020)
3WHOOP 3.00.441 study, n = 53 (κ = 0.44) · Miller et al. (2022)Maker funded
4Oura Ring Gen20.431 study, n = 53 (κ = 0.43) · Miller et al. (2022)
5Fitbit Sense0.421 study, n = 37 (κ = 0.42) · Schyvens et al. (2025)
5Fitbit Sense 20.421 study, n = 26 (κ = 0.42) · Lee et al. (2023)
5Samsung Galaxy Watch50.421 study, n = 22 (κ = 0.42) · Lee et al. (2023)
8Apple Watch Series 80.4152 studies, n = 88 (κ = 0.53; κ = 0.3) · Schyvens et al. (2025); Lee et al. (2023)
9Fitbit Charge 50.411 study, n = 39 (κ = 0.41) · Schyvens et al. (2025)
10Google Pixel Watch0.41 study, n = 30 (κ = 0.4) · Lee et al. (2023)
11WHOOP 4.00.371 study, n = 40 (κ = 0.37) · Schyvens et al. (2025)
12Oura Ring Gen30.351 study, n = 53 (κ = 0.35) · Lee et al. (2023)
13Polar Vantage V0.281 study, n = 53 (κ = 0.28) · Miller et al. (2022)
14Garmin Forerunner 245 Music0.251 study, n = 53 (κ = 0.25) · Miller et al. (2022)
15Withings ScanWatch0.221 study, n = 41 (κ = 0.22) · Schyvens et al. (2025)
16Garmin Vivosmart 40.211 study, n = 25 (κ = 0.21) · Schyvens et al. (2025)
17Apple Watch Series 60.21 study, n = 53 (κ = 0.20) · Miller et al. (2022)

Sleep vs wake (2-stage): epoch-by-epoch agreement (%)

Reference: Polysomnography (PSG). Higher is better. 5 studies; a device tested in more than one study shows its median.

1Oura Ring Gen3 (OSSA 2.0)91.7%1 study, n = 96 (91.7–91.8%) · Svensson et al. (2024)Maker funded
2Garmin Forerunner 245 Music89%1 study, n = 53 (89%) · Miller et al. (2022)
2Oura Ring Gen289%1 study, n = 53 (89%) · Miller et al. (2022)
2WHOOP (strap, 2.0 era)89%1 study, n = 12 (89% (κ 0.49)) · Miller et al. (2020)
5Apple Watch Series 688%1 study, n = 53 (88%) · Miller et al. (2022)
5Oura Ring (Gen2)88%1 study, n = 53 (0.88–0.89) · Chee et al. (2021)Maker funded
7Polar Vantage V87%1 study, n = 53 (87%) · Miller et al. (2022)
7Somfit87%1 study, n = 53 (87%) · Miller et al. (2022)
9WHOOP (auto-detect)86%1 study, n = 6 (86%) · Miller et al. (2021)Maker funded
9WHOOP 3.086%1 study, n = 53 (86%) · Miller et al. (2022)Maker funded

Sleep vs wake (2-stage): cohen's kappa

Reference: Polysomnography (PSG). Higher is better. 2 studies; a device tested in more than one study shows its median.

1Oura Ring Gen20.511 study, n = 53 (κ = 0.51) · Miller et al. (2022)
2WHOOP (strap, 2.0 era)0.491 study, n = 12 (κ = 0.49) · Miller et al. (2020)
3Somfit0.481 study, n = 53 (κ = 0.48) · Miller et al. (2022)
4Polar Vantage V0.441 study, n = 53 (κ = 0.44) · Miller et al. (2022)
4WHOOP 3.00.441 study, n = 53 (κ = 0.44) · Miller et al. (2022)Maker funded
6Garmin Forerunner 245 Music0.351 study, n = 53 (κ = 0.35) · Miller et al. (2022)
7Apple Watch Series 60.31 study, n = 53 (κ = 0.30) · Miller et al. (2022)

Total sleep time: pearson correlation

Reference: Research actigraphy. Higher is better. 1 study.

1Oura Ring (Gen2)0.861 study, n = 45 (r = 0.86) · Asgari Mehrabadi et al. (2020)
2Samsung Gear Sport0.591 study, n = 45 (r = 0.59) · Asgari Mehrabadi et al. (2020)

Total sleep time: mean absolute error (min)

Reference: Polysomnography (PSG). Lower is better. 1 study.

1Polar Ignite45.1 min1 study, n = 49 (MAE 45.08 min (overestimate)) · Budig et al. (2022)
2Garmin Forerunner 94584.6 min1 study, n = 49 (MAE 84.63 min (overestimate)) · Budig et al. (2022)

Head-to-head results within single studies

Devices tested side by side in one study, on the same people and protocol.

Ranked: higher is better

Multi-state (sleep stage + wake) epoch agreement vs PSG

Miller DJ, Sargent C, Roach GD (2022), Sensors. doi:10.3390/s22166317

53 adults, one lab night each · Reference: Polysomnography · 9 h in bed in a sleep laboratory, all six devices worn simultaneously

1Somfitκ 0.5265%
2Oura Ring Gen 2 Brand guideκ 0.4361%
3WHOOP 3.0 Brand guideκ 0.4460%
4Apple Watch S6 Brand guideκ 0.2053%
5Polar Vantage V Brand guideκ 0.2851%
6Garmin Forerunner 245 Brand guideκ 0.2550%

Ranked: higher is better

Two-state (sleep vs wake) agreement vs PSG

Miller DJ, Sargent C, Roach GD (2022), Sensors. doi:10.3390/s22166317

53 adults, one lab night each · Reference: Polysomnography · 9 h in bed in a sleep laboratory, all six devices worn simultaneously

1Oura Ring Gen 2 Brand guideκ 0.5189%
2Garmin Forerunner 245 Brand guideκ 0.3589%
3Apple Watch S6 Brand guideκ 0.3088%
4Somfitκ 0.4887%
5Polar Vantage V Brand guideκ 0.4487%
6WHOOP 3.0 Brand guideκ 0.4486%

Reported result (not ranked)

Sleep/wake performance (study conclusion)

Chinoy ED, et al. (2021), Sleep. doi:10.1093/sleep/zsaa291

34 healthy young adults, 3 nights · Reference: Polysomnography · Sleep lab, including a disrupted-sleep night

•Fitbit Alta HRAs good as or better than actigraphy
•Garmin Fenix 5S / Vivosmart 3Worse than actigraphy
•All devicesSensitivity ≥ 0.93, specificity 0.18–0.54

All published results

Every Sleep result in the database, including statistics that cannot be ranked. Download the full table as CSV for all rows. From the wearable accuracy database (v1.0.0).

Sleep stages (4-stage)9 studies

Device tested	Result	Reference	Sample	Study
Fitbit Sense	κ = 0.42Cohen's kappa	Polysomnography (PSG)	n = 37, Sleep laboratory, 1 night	Schyvens et al. (2025)Funding not reported
Fitbit Charge 5	κ = 0.41Cohen's kappa	Polysomnography (PSG)	n = 39, Sleep laboratory, 1 night	Schyvens et al. (2025)Funding not reported
Apple Watch Series 8	κ = 0.53Cohen's kappa	Polysomnography (PSG)	n = 62, Sleep laboratory, 1 night	Schyvens et al. (2025)Funding not reported
Oura Ring Gen3	76.0–79.5% (range across stages)Sensitivity	Polysomnography (PSG)	n = 35, Inpatient, 1 night	Robbins et al. (2024)Maker funded
Fitbit Sense 2	61.7–78.0% (range across stages)Sensitivity	Polysomnography (PSG)	n = 35, Inpatient, 1 night	Robbins et al. (2024)Other maker funded
Apple Watch Series 8	50.5–86.1% (range across stages)Sensitivity	Polysomnography (PSG)	n = 35, Inpatient, 1 night	Robbins et al. (2024)Other maker funded
Fitbit Sense 2deep sleep duration	Deep sleep −15 minMean bias (signed)	Polysomnography (PSG)	n = 35, Inpatient, 1 night	Robbins et al. (2024)Other maker funded
Apple Watch Series 8deep sleep duration	Deep sleep −43 minMean bias (signed)	Polysomnography (PSG)	n = 35, Inpatient, 1 night	Robbins et al. (2024)Other maker funded
Oura Ring Gen3	Not different from PSG for wake, light, deep or REMStudy finding (no single number)	Polysomnography (PSG)	n = 35, Inpatient, 1 night	Robbins et al. (2024)Maker funded
Oura Ring Gen3 (OSSA 2.0)	75.5% (light) to 90.6% (REM), per-stageEpoch-by-epoch agreement	Polysomnography (PSG)	n = 96, Home ambulatory PSG, up to 3 nights	Svensson et al. (2024)Maker funded
WHOOP (strap, 2.0 era)	64% (κ 0.47)Epoch-by-epoch agreement	Polysomnography (PSG)	n = 12, Sleep laboratory, 10-day protocol, 86 sleeps	Miller et al. (2020)Funding not reported
WHOOP (strap, 2.0 era)	κ = 0.47Cohen's kappa	Polysomnography (PSG)	n = 12, Sleep laboratory, 10-day protocol, 86 sleeps	Miller et al. (2020)Funding not reported

Fitbit Senseκ = 0.42
Cohen's kappa · vs Polysomnography (PSG) · n = 37, Sleep laboratory, 1 night
Schyvens et al. (2025) Funding not reported
Fitbit Charge 5κ = 0.41
Cohen's kappa · vs Polysomnography (PSG) · n = 39, Sleep laboratory, 1 night
Schyvens et al. (2025) Funding not reported
Apple Watch Series 8κ = 0.53
Cohen's kappa · vs Polysomnography (PSG) · n = 62, Sleep laboratory, 1 night
Schyvens et al. (2025) Funding not reported
Oura Ring Gen376.0–79.5% (range across stages)
Sensitivity · vs Polysomnography (PSG) · n = 35, Inpatient, 1 night
Robbins et al. (2024) Maker funded
Fitbit Sense 261.7–78.0% (range across stages)
Sensitivity · vs Polysomnography (PSG) · n = 35, Inpatient, 1 night
Robbins et al. (2024) Other maker funded
Apple Watch Series 850.5–86.1% (range across stages)
Sensitivity · vs Polysomnography (PSG) · n = 35, Inpatient, 1 night
Robbins et al. (2024) Other maker funded
Fitbit Sense 2Deep sleep −15 min
Mean bias (signed), deep sleep duration · vs Polysomnography (PSG) · n = 35, Inpatient, 1 night
Robbins et al. (2024) Other maker funded
Apple Watch Series 8Deep sleep −43 min
Mean bias (signed), deep sleep duration · vs Polysomnography (PSG) · n = 35, Inpatient, 1 night
Robbins et al. (2024) Other maker funded
Oura Ring Gen3Not different from PSG for wake, light, deep or REM
Study finding (no single number) · vs Polysomnography (PSG) · n = 35, Inpatient, 1 night
Robbins et al. (2024) Maker funded
Oura Ring Gen3 (OSSA 2.0)75.5% (light) to 90.6% (REM), per-stage
Epoch-by-epoch agreement · vs Polysomnography (PSG) · n = 96, Home ambulatory PSG, up to 3 nights
Svensson et al. (2024) Maker funded
WHOOP (strap, 2.0 era)64% (κ 0.47)
Epoch-by-epoch agreement · vs Polysomnography (PSG) · n = 12, Sleep laboratory, 10-day protocol, 86 sleeps
Miller et al. (2020) Funding not reported
WHOOP (strap, 2.0 era)κ = 0.47
Cohen's kappa · vs Polysomnography (PSG) · n = 12, Sleep laboratory, 10-day protocol, 86 sleeps
Miller et al. (2020) Funding not reported

28 more results in the database.

Sleep vs wake (2-stage)11 studies

Device tested	Result	Reference	Sample	Study
Fitbit Charge 5	> 90% (all devices)Sensitivity	Polysomnography (PSG)	n = 62, Sleep laboratory, 1 night	Schyvens et al. (2025)Funding not reported
Fitbit Sense	> 90% (all devices)Sensitivity	Polysomnography (PSG)	n = 62, Sleep laboratory, 1 night	Schyvens et al. (2025)Funding not reported
Withings ScanWatch	> 90% (all devices)Sensitivity	Polysomnography (PSG)	n = 62, Sleep laboratory, 1 night	Schyvens et al. (2025)Funding not reported
Garmin Vivosmart 4	> 90% (all devices)Sensitivity	Polysomnography (PSG)	n = 62, Sleep laboratory, 1 night	Schyvens et al. (2025)Funding not reported
WHOOP 4.0	> 90% (all devices)Sensitivity	Polysomnography (PSG)	n = 62, Sleep laboratory, 1 night	Schyvens et al. (2025)Funding not reported
Apple Watch Series 8	> 90% (all devices)Sensitivity	Polysomnography (PSG)	n = 62, Sleep laboratory, 1 night	Schyvens et al. (2025)Funding not reported
Oura Ring Gen3	≥ 95%Sensitivity	Polysomnography (PSG)	n = 35, Inpatient, 1 night	Robbins et al. (2024)Maker funded
Fitbit Sense 2	≥ 95%Sensitivity	Polysomnography (PSG)	n = 35, Inpatient, 1 night	Robbins et al. (2024)Other maker funded
Apple Watch Series 8	≥ 95%Sensitivity	Polysomnography (PSG)	n = 35, Inpatient, 1 night	Robbins et al. (2024)Other maker funded
Oura Ring Gen3 (OSSA 2.0)	91.7–91.8%Epoch-by-epoch agreement	Polysomnography (PSG)	n = 96, Home ambulatory PSG, up to 3 nights	Svensson et al. (2024)Maker funded
Oura Ring Gen3 (OSSA 2.0)	94.4–94.5%Sensitivity	Polysomnography (PSG)	n = 96, Home ambulatory PSG, up to 3 nights	Svensson et al. (2024)Maker funded
Oura Ring Gen3 (OSSA 2.0)	73.0–74.6%Specificity	Polysomnography (PSG)	n = 96, Home ambulatory PSG, up to 3 nights	Svensson et al. (2024)Maker funded

Fitbit Charge 5> 90% (all devices)
Sensitivity · vs Polysomnography (PSG) · n = 62, Sleep laboratory, 1 night
Schyvens et al. (2025) Funding not reported
Fitbit Sense> 90% (all devices)
Sensitivity · vs Polysomnography (PSG) · n = 62, Sleep laboratory, 1 night
Schyvens et al. (2025) Funding not reported
Withings ScanWatch> 90% (all devices)
Sensitivity · vs Polysomnography (PSG) · n = 62, Sleep laboratory, 1 night
Schyvens et al. (2025) Funding not reported
Garmin Vivosmart 4> 90% (all devices)
Sensitivity · vs Polysomnography (PSG) · n = 62, Sleep laboratory, 1 night
Schyvens et al. (2025) Funding not reported
WHOOP 4.0> 90% (all devices)
Sensitivity · vs Polysomnography (PSG) · n = 62, Sleep laboratory, 1 night
Schyvens et al. (2025) Funding not reported
Apple Watch Series 8> 90% (all devices)
Sensitivity · vs Polysomnography (PSG) · n = 62, Sleep laboratory, 1 night
Schyvens et al. (2025) Funding not reported
Oura Ring Gen3≥ 95%
Sensitivity · vs Polysomnography (PSG) · n = 35, Inpatient, 1 night
Robbins et al. (2024) Maker funded
Fitbit Sense 2≥ 95%
Sensitivity · vs Polysomnography (PSG) · n = 35, Inpatient, 1 night
Robbins et al. (2024) Other maker funded
Apple Watch Series 8≥ 95%
Sensitivity · vs Polysomnography (PSG) · n = 35, Inpatient, 1 night
Robbins et al. (2024) Other maker funded
Oura Ring Gen3 (OSSA 2.0)91.7–91.8%
Epoch-by-epoch agreement · vs Polysomnography (PSG) · n = 96, Home ambulatory PSG, up to 3 nights
Svensson et al. (2024) Maker funded
Oura Ring Gen3 (OSSA 2.0)94.4–94.5%
Sensitivity · vs Polysomnography (PSG) · n = 96, Home ambulatory PSG, up to 3 nights
Svensson et al. (2024) Maker funded
Oura Ring Gen3 (OSSA 2.0)73.0–74.6%
Specificity · vs Polysomnography (PSG) · n = 96, Home ambulatory PSG, up to 3 nights
Svensson et al. (2024) Maker funded

23 more results in the database.

Total sleep time6 studies

Device tested	Result	Reference	Sample	Study
WHOOP (strap, 2.0 era)	+8.2 ± 32.9 min (n.s.)Mean bias (signed)	Polysomnography (PSG)	n = 12, Sleep laboratory, 10-day protocol, 86 sleeps	Miller et al. (2020)Funding not reported
Fitbit Sense 2	−74.5 min (older adults)Mean bias (signed)	Polysomnography (PSG)	n = 19, 1 night	Searles et al. (2026)Funding not reported
Oura Ring	−75.5 min (older adults)Mean bias (signed)	Polysomnography (PSG)	n = 19, 1 night	Searles et al. (2026)Funding not reported
Oura Ring (Gen2)	−32.8 to −47.3 minMean bias (signed)	Polysomnography (PSG)	n = 53, Sleep laboratory, multiple nights, 3 sleep-opportunity conditions	Chee et al. (2021)Maker funded
Oura Ring	Group-level TST difference < 12 minStudy finding (no single number)	Polysomnography (PSG)	University sleep laboratory, clinical patients	Herberger et al. (2025)Independent
Oura Ring (Gen2)	r = 0.86Pearson correlation	Research actigraphy	n = 45, Home, 7 days	Asgari Mehrabadi et al. (2020)Independent
Samsung Gear Sport	r = 0.59Pearson correlation	Research actigraphy	n = 45, Home, 7 days	Asgari Mehrabadi et al. (2020)Independent
Garmin Forerunner 945	MAE 84.63 min (overestimate)Mean absolute error	Polysomnography (PSG)	n = 49, 1 night PSG + MSLT	Budig et al. (2022)Independent
Polar Ignite	MAE 45.08 min (overestimate)Mean absolute error	Polysomnography (PSG)	n = 49, 1 night PSG + MSLT	Budig et al. (2022)Independent

WHOOP (strap, 2.0 era)+8.2 ± 32.9 min (n.s.)
Mean bias (signed) · vs Polysomnography (PSG) · n = 12, Sleep laboratory, 10-day protocol, 86 sleeps
Miller et al. (2020) Funding not reported
Fitbit Sense 2−74.5 min (older adults)
Mean bias (signed) · vs Polysomnography (PSG) · n = 19, 1 night
Searles et al. (2026) Funding not reported
Oura Ring−75.5 min (older adults)
Mean bias (signed) · vs Polysomnography (PSG) · n = 19, 1 night
Searles et al. (2026) Funding not reported
Oura Ring (Gen2)−32.8 to −47.3 min
Mean bias (signed) · vs Polysomnography (PSG) · n = 53, Sleep laboratory, multiple nights, 3 sleep-opportunity conditions
Chee et al. (2021) Maker funded
Oura RingGroup-level TST difference < 12 min
Study finding (no single number) · vs Polysomnography (PSG) · University sleep laboratory, clinical patients
Herberger et al. (2025) Independent
Oura Ring (Gen2)r = 0.86
Pearson correlation · vs Research actigraphy · n = 45, Home, 7 days
Asgari Mehrabadi et al. (2020) Independent
Samsung Gear Sportr = 0.59
Pearson correlation · vs Research actigraphy · n = 45, Home, 7 days
Asgari Mehrabadi et al. (2020) Independent
Garmin Forerunner 945MAE 84.63 min (overestimate)
Mean absolute error · vs Polysomnography (PSG) · n = 49, 1 night PSG + MSLT
Budig et al. (2022) Independent
Polar IgniteMAE 45.08 min (overestimate)
Mean absolute error · vs Polysomnography (PSG) · n = 49, 1 night PSG + MSLT
Budig et al. (2022) Independent

Download all results (CSV)

Read before comparing

Ties in two-state agreement are ordered by Cohen's kappa.
One night per participant in a lab; home performance and disrupted sleep can differ (Chinoy et al. found worse performance on disrupted nights).
Rankings combine studies only when they report the same statistic against the same kind of reference, and still differ in protocol and population. Treat close values as ties.

Download the wearable accuracy table (CSV)

Every leaderboard row with study, DOI, reference standard and sample size. We'll email you the link.

FAQ

What is the most accurate sleep tracker?