Eight Sleep Pod 6

The system

A mattress cover and a hub. Water circulates through the cover from a 550 watt thermoelectric engine using three thermoelectric coolers per side, holding each half of the bed anywhere between 55 and 110 degrees Fahrenheit, independently for two sleepers. The Hub measures 16.54 by 7.87 by 6.50 inches and weighs 15.2 pounds, which is 846 cubic inches against 1,536 for the previous generation, and now fits under most standard bed frames. Optional additions are a Base that elevates the head and feet and carries a speaker, a Blanket that doubles the conditioned surface area, and a Pillow Cover, which requires a second Hub for independent regulation. Retail is 2,899 dollars plus a subscription.

What this company does that most do not

The Inspectorate makes a practice of recording candour where it finds it, because this category offers few opportunities. Eight Sleep publishes its internal research openly, with sample sizes, designs and stated limitations, on a blog anyone can read, which is why this file could trace every advertised figure to a source. It attributes every efficacy percentage to internal data in a footnote rather than implying independence. It carries the statement that the Pod is not a medical device prominently and site-wide. It describes itself as pursuing FDA clearance rather than implying it holds one, at a moment when several competitors are less careful. Its employee-authors disclose their equity in full in the peer-reviewed literature. Its chief executive apologised publicly and by name within a day of the 2025 outage. And it drew a distinction about pseudonymised data in its privacy policy that most companies quietly avoid. None of this changes the arithmetic. All of it is why the Inspectorate wrote nineteen entries about a product it could have dismissed in five.

What the sensors are

Eighteen piezoelectric elements, nine to a side, where the Pod 5 had one per side. Piezoelectric sensing here is ballistocardiography: the mechanical recoil of the body from each heartbeat and each breath is picked up as vibration and converted to a signal. From that the system derives heart rate, heart-rate variability, breathing rate, movement, snoring and sleep stage, with no wearable. The company states that the denser grid gives the system greater spatial awareness so it can separate one sleeper's biometrics from another's however close they sleep, and whether or not a child or pet joins them. That sentence is the entire public explanation. No method is disclosed, no accuracy figure exists under those conditions, and the previous generation's help documentation concedes that one person's movement could be attributed to the other's side.

What the Pod 6 changed

Half the Hub volume at 550 watts against 400. Temperature relief about twenty percent faster. Priming reduced from roughly twenty minutes to five, with the removable tank replaced by pouring directly into the Hub under LED guidance. The distilled-water requirement is gone; tap water is now acceptable. The maintenance insert is replaced by a liquid solution added twice a year. Water cord and USB cable are combined into one connector across all components. Every one of these answers a documented complaint about earlier generations, and taken together they are the clearest evidence in the file that the company listens to its owners.

Autopilot

The intelligence layer, at version 4.0 since April 2026. It takes sleep stage, heart rate, heart-rate variability, breathing, bedroom temperature, local weather, personal history, and since 4.0 daytime signals pulled from Apple Health and Health Connect including workouts, activity, recovery and alcohol. It adjusts temperature through the night and, each morning, delivers a plain-language account of what likely happened and how it responded. The company states the system is fully personalised after about seven nights. Fully personalised is defined nowhere, no convergence criterion is published, and the same page says it only gets better with time. Whether a user can override it, and whether it learns from overrides, is stated in no document the Inspectorate could find.

The open question at the centre of Autopilot

Autopilot has not yet been compared against a manually programmed temperature schedule. No study, internal or external, published or preprint, has run that comparison. The one head-to-head that exists compared Autopilot 2.5 against Autopilot 2.0, which is a real test of an upgrade. The conference abstract on stage-based adjustment compared sleeping on the Pod against not sleeping on the Pod, which cannot separate the algorithm's contribution from the cooling hardware's. The Inspectorate raises this not as a criticism of the algorithm, which may well be doing exactly what the company believes it does, but because it is the most interesting claim in the product and nobody has yet measured it. A company with twelve PhDs and a fleet this size is better placed to answer it than anyone alive.

What the independent science establishes

A great deal, and it is genuinely favourable to the mechanism. Mild skin-temperature manipulation of about four tenths of a degree, within the comfortable range and without changing core temperature, measurably alters polysomnography-scored sleep: it suppresses wakefulness and increases slow-wave sleep. In healthy elderly participants a 0.4 degree increase nearly doubled slow-wave sleep and reduced the probability of early-morning awakening from 0.58 to 0.04. The distal-to-proximal skin gradient predicts sleep onset latency better than core temperature, better than heart rate, better than melatonin onset and better than how sleepy a person says they are. The best-controlled multi-site study of a cooling sleep surface, 72 people across three cities with full polysomnography, found an increase of 7.5 minutes of deep sleep across a 7.5 hour night and a heart rate 2.36 beats per minute lower.

What the independent science does not establish

Three things the Inspectorate regards as decisive. First, proximal and distal warming have opposite signs: warming the torso reduces wakefulness and increases deep sleep, while warming the extremities increases wakefulness and light sleep at the cost of deep and REM sleep. A surface that conditions the whole body does not distinguish them. Second, the founding work in this field states that circadian thermoregulation initiates sleep but does not appear to play a major role in its maintenance, which is the premise the all-night adaptation rests on. Third, thermoregulation is largely abolished during REM sleep; a sleeper in REM cannot defend body temperature, and the only response available is waking. That is a mechanistic argument for small, slow, conservative changes, and no independent trial has tested stage-triggered thermal actuation against a sham.

The one randomised independent test

Stevenson, Suppiah, Mundel and Driller, published in Clocks and Sleep in October 2025. Randomised, counterbalanced crossover, 34 healthy adults, seven nights with the temperature system active and seven with it disabled, funded by La Trobe University with no declared conflicts. The subjective results were strong and significant across the board: calmness, ease of falling asleep, ease of waking, feeling refreshed, sleep satisfaction, thermal sensation and thermal comfort, with weekly sleep quality improving at an effect size of 0.92, which is large by any standard. The objective measures did not move: total sleep time, wake after sleep onset, sleep onset latency, sleep efficiency, heart rate, heart-rate variability and respiratory rate all returned no significant difference. The authors noted that expectancy may account for part of the subjective gain. The Inspectorate records the whole of that result in both directions, and adds the context that makes it fair: the largest effect ever verified by polysomnography for any cooled sleep surface, in the best-controlled study in this field, is seven and a half minutes of additional deep sleep across a whole night. An effect of that size is genuinely difficult to detect in 34 people over seven nights. The null is a real finding and it is not proof of absence.

The four percentages, and what they measured

Eight Sleep publishes the studies behind its headline numbers openly, with sample sizes, methods and stated limitations, on a blog anyone can read. Very few companies in this category publish anything at all, and this file could trace every advertised figure to a source only because the company made that possible. What the sources show is that the published numbers are ceilings rather than averages. Up to 34 percent more deep sleep is the top of a correlation among the most sleep-deprived participants in an uncontrolled study of 310 existing customers measured by a competitor's ring; the average across all 310 is not published. Up to 44 percent less time to fall asleep and 32 percent better sleep quality come from one uncontrolled retrospective questionnaire of 50 people; under a randomised control condition the sleep-onset difference was one minute and not significant. Reduce snoring by up to 45 percent is the figure at 15 degrees of head elevation, the most aggressive setting, in an internal study of 39 people, and it requires the Base; at 6 degrees the figure is 31 percent. All four carry a footnote attributing them to internal study data, which is to the company's credit. Publishing the averages and the confidence intervals beside the ceilings would cost very little and would make these claims unimpeachable.

Clinical-grade accuracy

The phrase appears throughout the marketing. On the product pages it carries a footnote, and the footnote resolves not to a study but to the words Autopilot membership required. The underlying data are real and internal: heart rate against electrocardiogram across 163 subjects and 474 nights returned a correlation of 0.996 with mean absolute error under one beat per minute, which is genuinely excellent; heart-rate variability returned a mean absolute error of 5.7 milliseconds, which is a material fraction of a signal that typically runs between 20 and 80; sleep staging is reported at 78 percent accuracy with no Cohen's kappa published, which is below the technician agreement the company itself cites as the benchmark. All of it was generated on a Pod 3 with two sensors. The Pod 6 has eighteen and has been validated by nobody.

On measuring your own intervention

A methodological point that applies to this whole category rather than to Eight Sleep alone. The independent literature on consumer sleep staging is consistent: non-electroencephalographic devices detect sleep well and wake poorly, and four-stage discrimination remains unreliable, with deep-sleep sensitivity between 0.53 and 0.68 and positive predictive value between 0.44 and 0.58. Accuracy degrades in exactly the disturbed sleepers who most want help. For ballistocardiography specifically, one clinical-population validation found total sleep time overestimated by nearly three hours and staging that did not function at all. Several of Eight Sleep's internal studies measure the effect of the Pod using the Pod's own sensors, which is standard practice across this industry and is nonetheless a problem the industry has not solved, because the measurement error is larger than the largest effect in the independent literature. The Inspectorate records this as the reason it weights the one polysomnography-referenced independent trial so heavily, and as the strongest argument for the company to fund an external one.

On menopause

The company has studied nocturnal temperature regulation in peri- and post-menopausal women and reports a 56 percent reduction in hot flashes across a 14-night protocol in 98 women. The work is a preprint and has not passed peer review. Its methods are the company's strongest: ingestible core-temperature telemetry, skin thermistors at three sites, and urinary hormone confirmation of menopausal status. Its weaknesses are serious: the primary outcome is self-reported, 39 of the 98 were not randomised but received a fixed order, participants were recruited from existing Pod owners, and the standard deviations on two of the three reported outcomes are four times the size of the effect. Set against it, the 2023 position statement of the North American Menopause Society lists cooling techniques among interventions that are not recommended, and the one polysomnography study of this exact population found that hot flashes did not predict objective sleep efficiency at all, while 53 percent of the women had undiagnosed apnea, restless legs or both. The Inspectorate records that a cooling product which displaces a sleep-apnea workup is a net clinical harm, and that this is the single most important caution in the file.

Regulatory standing

Eight Sleep holds no FDA clearance and no Abu Dhabi HTA clearance for any function. It describes itself as pursuing both for sleep apnea, present tense, and a trial with Abu Dhabi's Department of Health began in August 2026 with no results published. The site-wide disclaimer that the Pod is not a medical device and is not intended to diagnose, treat, cure or prevent any disease is correct and prominent. The Inspectorate records the position exactly as it stands today, and notes that snoring reduction is not treatment for obstructive sleep apnea and must never be presented as such.

The warranty, and what the documents say

The product page displays a five-year warranty. The warranty document names the Eight Sleep Pod 6 in the clause providing two years. Five years is available only by purchasing Autopilot Enhanced or Elite at checkout, and the same document states that warranty coverage requires an active Eight Sleep membership during the applicable coverage period for both standard and extended warranties. Read plainly, ending the subscription puts coverage of either length at risk. Downgrading collapses five years to two; upgrading does not restore it. Coverage runs from delivery rather than first use, is non-transferable, and replacements do not restart the clock. Consequential damages are expressly disclaimed, so a cover that leaks onto a mattress leaves the mattress as the owner's loss. Disputes go to confidential binding arbitration in New York with a class-action waiver, and total liability is capped at the greater of twelve months of payments or one hundred dollars. None of this is unusual for connected consumer hardware. The Inspectorate records it because a buyer reading the product page would reasonably expect five years unconditionally, and stating the two-year term there with the upgrade offered alongside it would cost the company nothing and close the gap entirely.

What stops when the payment stops

The manual temperature dial survives and each side can still be set by hand. What disappears is the temperature schedule, automatic adjustment, the vibration and thermal alarms, all sleep and health reports, Away Mode, snoring detection and snoring mitigation. The Terms of Service state that an active subscription is required to access core App features and full Pod functionality including thermoregulation, biometric tracking and AI-powered personalisation, and that without one the Pod may continue to provide limited basic temperature control. Autopilot is required for the first twelve months of ownership, a term that appears in the help centre and nowhere in the contract. What happens to years of accumulated biometric history on cancellation is addressed in no document the Inspectorate could find.

Privacy, where the company is better than its own summary

Eight Sleep does something here that almost no company in this category does: it states explicitly that pseudonymised data retains indirect identifiers and is not considered deidentified. That is a real distinction, honestly drawn, and it is to the company's considerable credit. Biometric data, the heart rate and breathing and variability that people most worry about, is excluded from advertising and remarketing use. The gap is in the summary rather than the substance. The binding Privacy Notice states that the company does sell and share identifiers, commercial information and internet activity for marketing, advertising and analytics purposes. The help centre states that the company does not sell personal information to third parties. Both cannot be true as written, and correcting the help centre to say what is actually the case, that biometrics are not sold, would leave the company better represented than it currently is. One further note: the encryption commitment lives in a blog post rather than in the binding notice, which contains no encryption language and says instead that information sent electronically may not be secure while in transit.

Connectivity, and what the object is without it

The Terms of Service state that the Pod requires a Wi-Fi internet connection and the App to operate, and that the Services depend on cloud connectivity. There is no documented offline or local-only mode. The Hub carries one physical button and its only documented function is pairing. In October 2025 a cloud provider outage left beds unresponsive, with one side reported stuck at 110 degrees and bases stuck upright; the chief executive apologised publicly within a day and promised outage-proofing, and as of January 2026 no offline mode had shipped or been discussed again. For a device that costs nearly three thousand dollars and is designed to last, the Inspectorate regards mandatory cloud dependency as the most consequential design decision in the product.

Who else is doing this

Nobody is doing the same thing better, and that is a genuine finding. Eight Sleep is the only consumer system that closes a loop on two axes, thermal and postural, from physiological sensing. The nearest rival on mechanical sophistication controls four hundred springs individually and costs between fifty and a hundred and ten thousand dollars, has published no evidence of any kind, and its cooling does not ship until late 2026. The company with the best sensing evidence, two peer-reviewed polysomnography validations, controls temperature on a timer rather than on the body and entered Chapter 11 in June 2026. The two best-known water and air systems do not sense the body at all, and one says so plainly in its own marketing. Against the medical standard the gap remains wide: targeted temperature management in hospital senses core temperature invasively, holds it to two tenths of a degree, readjusts every two minutes and limits the gradient between patient and water as a safety principle. No consumer sleep company publishes a control-accuracy specification. Not one.

How the tier was computed

Ninety-one sub-criteria were scored at the published anchors, where 3 meets the institutional standard, 4 exceeds it and 5 sets it. Circle I passed on all seven binary tests. Three sub-criteria were recorded N/A: culinary integrity, regenerative practice and cultural authenticity, none of which has a subject in a bedroom device. The honest first pass returned an institutional score of 3.42 with a floor of 3.00 at Stewardship, which is Finalist: the floor cleared the requirement for One Bio Star and the institutional score fell eight hundredths short of the 3.5 required. The principal ruled One Bio Star on the strength of months of nightly use, and seven scores were revised on that evidence, each carrying the ruling in its note: sensory burden, usability, real-life integration, repeatability, nervous-system stability, component selection and operational learning. Five of the seven are criteria on which direct nightly use is the best evidence available and on which the Inspectorate had scored an absence of published data rather than an observed failure. The revised file computes an institutional score of 3.50 with the floor unchanged at 3.00, which is One Bio Star at the anchors. Two Bio Stars requires a floor of 3.5, a score of 4.0 and six Circles at 4.0 or above; this file has two. The distance is almost entirely documentary rather than mechanical, and a single independent trial would close most of it. The Inspectorate regards this as a file with an unusually short road to its second star.