Finding the Best Sleep Tracker | karpathy

The Start of a Sleep Improvement Project

On March 24, 2025, the author happened to watch Bryan Johnson's YouTube video "How I FIXED My Terrible Sleep - 10 Habits". This video prompted the decision to improve sleep quality. But to improve sleep, first you need to figure out how to measure it. While Bryan Johnson uses Whoop, the author was using an Apple Watch with the AutoSleep app, and had previously used Oura. Additionally, the newly released 8Sleep Pod 4 Ultra was on order.

So after some deliberation, a project comparing 4 major sleep trackers was launched.

2 Months of Data Collection

Over 2 months, every morning the following data was recorded:

• Sleep stages: Light, REM, Deep, Awake
• Heart rate data: Resting Heart Rate (RHR), Heart Rate Variability (HRV)
• Sleep scores: Scores provided by each app

The collected data was organized in a spreadsheet, with an example below.

Sleep Tracker Evaluation

After analyzing 2 months of data, each device performed as follows:

1. Oura and Whoop: Rated "Tier 1." Provided high-quality sleep tracking, and scores generally matched morning mood well.
2. 8Sleep: Decent level, but below Oura and Whoop.
3. Apple Watch + AutoSleep: Scores appeared nearly random, with low correlation to morning mood.

Data Analysis: Score Distributions

Below is a histogram showing the distribution of scores provided by each device over 2 months.

• AutoSleep and 8Sleep: Scores tend to skew excessively high, frequently giving scores close to 100.
• Whoop: Gives relatively high scores, but reaching 100 is easy.
• Oura: The most strict scoring, showing a Gaussian distribution (normal distribution). The highest score was 92, leaving room to keep striving for better sleep.

"Oura's scores don't easily saturate and provide a dynamic range, which is very satisfying."

Inter-Device Correlations

Analyzing score correlations between devices yielded the following results:

• Whoop and Oura: Highest correlation (0.65).
• Whoop and AutoSleep: Almost no correlation (0.14). AutoSleep looks like a "noise generator."

"If Whoop is a good device, AutoSleep is generating nearly random numbers."

Heart Rate Data

Analysis of Resting Heart Rate (RHR) and Heart Rate Variability (HRV) data:

• 8Sleep doesn't directly measure heart rate, so accuracy is lower.
• The remaining devices (Oura, Whoop, AutoSleep) provided heart rate data consistently.

Exercise and Sleep Correlation

Over 2 months, the author averaged 30 minutes daily of Zone 2 cardio and twice-weekly HIIT (High-Intensity Interval Training). The results:

• Resting Heart Rate (RHR): Decreased by about 3 bpm (51 to 48 bpm)
• Heart Rate Variability (HRV): Increased by about 5 (49 to 54)

"Seeing exercise reflected in the data is really cool."

The Importance of Sleep

The author emphasizes that sleep scores are closely connected to daily productivity.

"On days with low sleep scores, I lack motivation, courage, and creativity. On days with high scores, I can work 14 hours without noticing the time."

Additionally, sleep quality is determined not by a single night's sleep but by accumulated sleep debt over recent days.

Sleep Tracker Recommendations

The author summarizes the pros and cons of each device with the following recommendations:

1. Oura:

   • Pros: Excellent sleep tracking, intuitive app interface, dynamic scoring.
   • Cons: Ring form factor can be uncomfortable (washing hands, cooking, exercising).
   • Rating: 10/10

2. Whoop:

   • Pros: Convenient wrist wear, can be worn while charging, community features.
   • Cons: App is somewhat complex, scores saturate easily.
   • Rating: 9.5/10

3. 8Sleep:

   • Pros: Decent level of sleep tracking.
   • Cons: Scores not reliably trustworthy.
   • Rating: 6/10

4. AutoSleep:

   • Cons: Scores appear nearly random.
   • Rating: 2/10

Conclusion

"Sleep is an important factor that determines quality of life. Whoop and Oura are the top choices for sleep tracking. If you want to improve your sleep, choose one of these two."

Finally, the author asks readers to let him know via X (formerly Twitter) if there are any sleep trackers he missed.