← Longevity & Supplement Guides
Biological Age Tests, Honestly: What the 2026 Research Found
The report says you are thirty-four. Your passport says forty-one. For a day or two this is the best news you have had all year, and then a quieter question arrives: what, exactly, was measured?
Biological-age testing has moved from the laboratory to the checkout page, and 2026 has turned out to be an unusually honest year for it. Within a few months the field produced one study that made these tests look better than every cheap alternative, and another line of work arguing they are not yet fit to be read by one person about one body. Both are defensible. The gap between them is where the useful thinking lives.
The number that arrives in an email
Most consumer biological-age products work from DNA methylation — chemical tags that sit on top of your DNA and shift with age, environment and behaviour. An algorithm trained on thousands of people reads a few hundred of those sites and returns a figure.
The generations matter. First-generation clocks were trained to guess chronological age, which is a strange goal: succeed perfectly and you have built an expensive calendar. Second-generation clocks were trained on mortality and health outcomes instead. Third-generation clocks, of which DunedinPACE is the best known, ask a different question again — not how old you are, but how fast you are ageing, expressed as biological years per calendar year.
That distinction deserves more attention than the marketing gives it. A static age is a snapshot. A pace is a slope, and a slope is the part you can plausibly bend.
Fourteen markers, one fair race
In March 2026, researchers working with the Berlin Aging Study II published the comparison the field had been missing: fourteen consensus biomarkers of ageing, measured in the same older adults, raced against each other to see which one tracked mortality. The analysis covered 1,083 participants with complete data, aged 60 to 80 at baseline, followed for an average of 7.4 years.
The list was deliberately unglamorous alongside the genomics — inflammatory markers such as IL-6 and CRP, IGF-1, blood pressure, muscle mass, hand grip strength, gait speed, the Timed-Up-and-Go, standing balance, cognitive health, a frailty score, and the epigenetic clocks. Five markers reached statistical significance: grip strength, IL-6, standing balance, cognitive health, and DunedinPACE. Of those, the pace-of-ageing clock showed the strongest and most consistent association with mortality.
That is a real result, and a mild embarrassment for the sceptics. In a fair race, the expensive methylation test beat the cheap physical tests on their own terms.
Evidence
What the best ageing panel can and cannot resolve
One readout, drawn with its error bars estimated years
How well the best panels sort who dies (C-index)
Which of the fourteen markers actually tracked mortality?
Reached significance: DunedinPACE (strongest and most consistent), hand grip strength, IL-6, standing balance, cognitive health.
Did not, in this cohort: IGF-1, methylation-derived GDF15, CRP, muscle mass, gait speed, Timed-Up-and-Go, frailty phenotype, blood pressure, and a first-generation epigenetic clock.
Muscle mass still earned a place in the three-marker minimal model, which is a useful reminder that "significant on its own" and "useful in combination" are different questions.
Even the winner is a blunt instrument
Then the same paper delivers the deflating detail. The full model, using all fourteen biomarkers, reached a C-index of 0.65. A minimal model using only three — muscle mass, standing balance and DunedinPACE — reached 0.63. A C-index of 0.50 is a coin flip and 1.0 is perfect foresight. So the best available panel of ageing biomarkers, in a well-characterised cohort, separates who died from who did not somewhat better than chance and nowhere close to certainty.
This is ordinary for population science and ruinous for personal prophecy. The panel is informative about groups of people and vague about you.
Where the honest doubt lives
The reliability literature is blunter still. A December 2025 review in Epigenomics, examining the limits of these clocks as personal biomarkers, noted that technical replicates of the same sample have produced age estimates differing by as much as nine years, that median absolute error for first-generation clocks has been reported at 3.6 years or higher, and that methylation itself shifts on a timescale of minutes to hours. Set that against a routine glucose assay, where acceptable bias is measured in low single-digit percentages, and the ambition of reading your own number to one decimal place starts to look decorative.
The clocks also disagree with each other. A Nature Communications paper published in December 2025 ran fourteen widely used clocks against 174 incident disease outcomes and all-cause mortality in 18,859 people across ten years of follow-up. Second- and third-generation clocks clearly outperformed the first generation — but no single model emerged as uniformly superior. An eBioMedicine editorial in February 2026 made the same point more diplomatically, while flagging that most clocks rest on correlation rather than demonstrated mechanism, and that real questions remain about reliability in commercial wellness settings.
Two of the three survivors are free
Now look again at that minimal three-marker model: muscle mass, standing balance, DunedinPACE. Two of the three cost nothing. You can time how long you hold a one-legged stand before breakfast tomorrow. You can notice whether the shopping feels heavier this year than last — and grip strength, the fourth significant marker in the Berlin comparison, needs a cheap dynamometer at most.
This is not an argument against testing. It is an argument about order. If two of the three most useful ageing signals in a fourteen-marker race are things you can measure in your own kitchen, that is where a reasonable person starts — and the thing that moves them is the tissue you build by using it.
How to hold the number
If you do buy a test, hold it the way you hold a bathroom scale: as one noisy reading of something real, not a verdict handed down.
Repeat with the same provider and the same method, because comparing across clocks compares algorithms as much as bodies. Read the direction across years rather than the digits. Expect whatever moves the number to be dull — activity, sleep, body composition, glucose, not smoking — which is the pattern the recent literature keeps landing on. Be sceptical of any product claiming to reverse your age by a decade or describing itself as clinically proven. And keep the category straight: this is a wellness estimate, not a medical test, and it diagnoses nothing. Our wider guide to which biomarkers are worth your attention puts it in context, as does the honest accounting of what a wearable can and cannot measure.
The bottom line
The 2026 evidence is genuinely two-sided, and it is more interesting for that. Pace-of-ageing clocks are the best single biomarker of ageing yet tested in a fair race, and they remain far too noisy to tell any individual their future. Both facts survive.
Which leaves the position I keep returning to: use numbers to correct fantasy, not to replace experience. A biological-age report is worth reading if it makes you honest about a slope you had been ignoring. It is worth ignoring the moment it becomes the thing you do instead of standing on one leg, carrying the shopping, and going to bed.
