Fitbod published a new 1RM formula.

First, what a 1RM estimate is

Your one-rep max is the heaviest weight you can lift for a single rep. Most lifters almost never test it. It needs a spotter, it costs a training day, and it is a bad idea in week one of a block.

So we estimate it instead. Take a hard set, run it through an equation, get a number. Epley, the formula behind the LIFTAG 1RM calculator and the estimate in the app, is 1RM = weight × (1 + reps / 30). A set of 100 kg for 5 comes out at 116.7 kg. Brzycki, the other one everybody quotes, says 112.5 kg.

That number is a reference point for programming. It is not a meet attempt, and it only means anything if the set was close to failure.

What Fitbod actually did

In March 2026, Thiago Marzagao, head of data science at Fitbod, posted a preprint called “A Weight-Dependent Equation for One-Repetition Maximum Estimation”. It is on SportRxiv and on arXiv.

The dataset is the part worth paying attention to. 303,494 near-failure sets across 388 exercises, pulled from Fitbod app logs. The lab studies behind Epley and Brzycki typically ran a few dozen people through a bench press. This one covers machines, cables and dumbbells as well as barbells.

The complaint driving the paper is simple. Every classic equation uses one fixed conversion from reps to 1RM, no matter what is in your hands. That conversion was mostly derived on the bench press. On light isolation work it misses badly.

The formula, in plain language

Written out, in kilograms:

• k = -2.55 + 4.58 × ln(weight)
• estimated 1RM = weight × (1 + (reps − 1)^0.85 / k)

Put that next to Epley: 1RM = weight × (1 + reps / 30). Epley always divides by 30. Brzycki works out to roughly 36. This paper makes the divisor move with the load.

At 10 kg the divisor is about 8. At 70 kg it is about 17. At 150 kg it is about 20. A small divisor means each extra rep pushes the estimate up harder. In gym terms: ten reps on a light curl tells you much more about your single than ten reps on a heavy squat does.

The paper reports that the weight-dependent part accounts for roughly 91% of the improvement and the 0.85 exponent on reps for the other 9%. The exponent was fixed before the fit; the two coefficients were grid-searched.

One practical detail. It is calibrated in kilograms, and Fitbod logs dumbbells per hand. Feed it one dumbbell, not the pair.

Same reps, different bar

Here is what the three equations say about the same sets. All numbers in kilograms, dumbbells per hand, rounded to one decimal.

On the heavy barbell sets all three land within a few kilos of each other, which is less than one plate change. On a 13 kg curl for 10 the new equation is almost 5 kg higher than both classics. That gap is the entire argument of the paper.

What the study does not prove

They never tested a real one-rep max. Not one. The check was internal consistency: if the same person does several near-failure sets inside a two-week window, a good formula should predict roughly the same max from all of them. That is a fair sanity test. It is not the same as walking up to a loaded bar and lifting it.

• “Near failure” was inferred from the logs, using AMRAP flags and rep drop-off. Nobody watched those sets.
• It is one company’s self-logged app data. People round up and misremember.
• The sample is roughly 80% male and mostly aged 25 to 39.
• It is a preprint. No peer review yet, and no DOI.
• The author works for Fitbod. He discloses it in the paper. It is still a conflict of interest.
• Weight is standing in for exercise type. The equation cannot see a paused bench versus a bounced one, two minutes of rest versus five, or how much the last rep ground.
• Like every 1RM equation, it gets shakier as the reps climb.

Credit where it is due. The filters, the coefficients and a genuinely self-critical limitations section are all written down where anyone can check them. Most apps sitting on a dataset that size publish a marketing claim instead. This is better than that. It is also not landmark, not definitive, and not the new standard. It is a preprint with a good idea in it.

What this changes for your training

For barbell work, nothing. On a top set of 3 to 5 on the bench or the squat, every one of these equations lands inside the same rounding error. You would load the same bar either way.

For isolation work, it is a useful reminder that the estimated max on a lateral raise was never worth much in the first place. Nobody programs percentages off a curl max.

The habit that actually moves the number has not changed. Log the set, keep the standard, add one thing at a time. Progressive overload and an honest RPE when the set was a grind will tell you more than any argument about divisors.

What LIFTAG is doing with it

Nothing yet. The app and the calculator still default to Epley, and the calculator shows Brzycki, Lombardi, Mayhew, O’Connor, Wathen and Lander next to it so you can see the spread for yourself. All seven are listed with their equations on the formula breakdown.

We are not swapping the math on a preprint. If independent groups test it against real, measured maxes and it holds up, we will look again. The underlying idea, that one fixed conversion cannot cover a squat and a lateral raise, is plausible and lines up with older work on exercise specificity. It is worth watching.

What we will not do is quietly change the formula and let your whole history shift underneath you.

Sources

• Marzagao, T. “A Weight-Dependent Equation for One-Repetition Maximum Estimation.” SportRxiv preprint, posted 18 March 2026: sportrxiv.org
• The same preprint on arXiv, identifier 2603.17495: arxiv.org
• Epley and Brzycki figures in the table were computed from the published equations we list on the LIFTAG 1RM calculator.

Links checked 21 September 2026. The preprint has not been peer reviewed.