Method

Data status

All 80 scores shown are placeholders: 66 read from EveryMac, which may be used only to cross-check, and 14 from Primate Labs (Geekbench Browser), pending permission. Its terms do not allow publishing benchmark results without permission, so collection from it is paused. Placeholders are unverified values; each will be replaced by a sourced value or removed before launch. 319 of 399 score values are missing. A missing value is shown as “—”, never as zero, and is never interpolated. No scores yet for iPhone base, iPad Air, iPad, iPad mini, MacBook Pro, Mac mini, iMac and Mac Studio / Mac Pro.

How scores are averaged

Measured average (62 scores)

The Geekbench 6 average that Primate Labs publishes for the device. Averages are not recomputed from individual results. When a chart lists several configurations, the one named on the device page is used; configurations are never mixed. The sample size is shown when the chart states it.

Early result, marked ≈ (2 scores)

For a device not yet on a chart: the median of individual Geekbench 6 results, each linked, until a chart average exists.

Converted, marked ≈ (16 scores)

For a device that cannot run Geekbench 6: its average in an older Geekbench version times a conversion factor.

GPU scores use the Metal API. Devices without Metal have no GPU score.

Which model represents a generation

iPhone flagship

Each year’s top model: original, 3G, 3GS, 4, 4S, 5, 5s, 6, 6s Plus, 7 Plus, 8 Plus, XS, then Pro.

iPhone base

The standard numbered model each year from the first year with more than one model.

iPad Pro

One entry per chip: the first model with each chip; the smaller size when two sizes launch together.

iPad Air

One per generation; the 11-inch model from M2 on.

iPad

The base iPad per generation.

iPad mini

One per generation.

MacBook Air

13-inch, base configuration per generation.

MacBook Pro

One per processor generation: the first 13-inch base model (the 15-inch in 2007), the 14-inch base model from M3.

Mac mini

Base configuration per generation.

iMac

Smaller-screen base configuration per generation.

Mac Studio / Mac Pro

The top configuration of the top desktop of each generation: Mac Pro until 2019, then Mac Studio with the Ultra chip.

Conversions

A factor comes from bridge devices with averages in both an older Geekbench version and a newer one. Each bridge gives a ratio, the newer score over the older one; the factor is their mean and the spread their lowest and highest value. A chained factor steps through each version in between. Scores converted in one step are rounded to whole points; scores converted through two or more versions keep two significant figures. No factor has a recorded spread yet, so converted scores show no range.

What is estimated

8 devices have converted scores, launched 2007 to 2014. Scores converted from Geekbench 2 (devices launched 2007 to 2010) give the order of magnitude only. Early results: iPhone 18 Pro. Estimated scores carry “≈” and are drawn dashed on the chart.

How verdicts are computed

A claimed percentage gain becomes a factor of one plus that percentage; a claimed multiple is used as it is. The measured ratio is the device’s score over the score of the device Apple compared it with, on the matching measure: single-core, multi-core or GPU (Metal). For a CPU claim that names neither, the better of single-core and multi-core decides.

Measured over claimed within ±5% is “= Matches”, above it “> Exceeds”, below it “< Falls short”. The band is the same for “up to” claims and plain ones. A verdict is provisional when either score is an early result and estimated when one is converted. Claims Geekbench cannot test, such as Neural Engine speed, battery life or graphics in games, get no verdict.

Known limits

• Geekbench runs for seconds, so it measures burst speed. Sustained work, where a phone or a fanless laptop slows down as it heats up, is not captured.
• Geekbench 7 has replaced Geekbench 6 on Geekbench Browser. It uses a new baseline, so its scores cannot be mixed with these; this site stays on Geekbench 6, which covers far more devices.
• Converted scores (devices launched 2007 to 2014) are estimates: alternative conversion chains differ by up to 6%, and the error grows with each version stepped back.
• Averages pool results that users upload across operating system versions, and move as results accumulate.