Global Reflex Statistics

Live, anonymized statistics from GGReflex reflex tests. Every number below is computed from real test results, one best score per player per game, and refreshes continuously. No sign-up data and no tracking of individuals. Nothing is weighted or adjusted to represent a population either: anyone can take the tests from anywhere, so these are the reflexes of whoever turned up. The sample behind each median is published further down the page, broken out by country and by device, so a figure you quote can be quoted with the people it came from.

Methodology

Scores are measured client-side using high-resolution timers (performance.now()), with the clock starting when the stimulus is issued and stopping when the browser reports the input. Every measurement therefore contains the participant's display and input latency as well as their reaction, and a web page cannot separate the two. We do not attempt to correct for it, and we will not quote a figure for it: we are not aware of a defensible published estimate of the total overhead a browser test adds, and inventing one would defeat the purpose of this page.

The components that can be stated exactly are these. A display refreshing at 60 Hz redraws every 16.7 ms, so a stimulus ready in software may wait up to one frame to become visible — roughly half a frame on average for an event that can occur at any time. At 144 Hz that worst case falls to 6.9 ms. A mouse polling at 125 Hz may hold an input for up to 8 ms before reporting it; at 1000 Hz, about 1 ms. Touch input is slower again. These figures bound part of the overhead; they do not sum to a total, and the remainder — browser and compositor scheduling — we cannot quantify honestly.

The practical consequence is that these distributions are web-measured distributions. They are directly comparable with each other and with other browser-based tests, and they are not comparable with laboratory instrumentation, which reports faster numbers because it removes hardware the participant is still using.

Before aggregation we keep one best score per player and rate-limit submissions per address. We do not apply a plausibility filter per test, because we have no defensible threshold to apply: the score left behind by someone who started a reaction test and walked away is in this sample, and so is whatever an auto-clicker produced. The medians and percentiles above are computed from every score anyway — a median is barely moved by a handful of absurd values, which is why this page reports medians rather than averages, always with the number of players behind them.

Where the sample allows it, a test is published once per device rather than once overall, because on some of these tests the input device is a large part of what is being measured. Click speed is the clearest case: in a snapshot taken in August 2026 the median on touch screens was 17.2 clicks per second across 48 players against 7.2 across 136 on desktops, and 27 of the 35 scores above the fastest recorded single-finger clicking came from touch devices. Nothing is being cheated — a touch screen registers several fingers at once and a mouse button does not, so the two are not the same task, and one median across both describes nobody. The same effect runs the other way on tests where touch latency costs you: in that snapshot the reaction-time median was 262 ms across 226 desktop players against 321 ms across 79 on mobile. We split whenever both groups reach the 20-player floor, and never when only one does, since a chart labelled with one device while the other's players are missing would mislead more than the mixture it replaced. The figures in this paragraph are a dated illustration of why we split; the live split for every test is on the cards above. Scores that arrived without a device recorded are counted in the combined figures and left out of the split.

Within the desktop figure the same caveat applies to technique rather than hardware. Jitter, butterfly and drag clicking are established ways of getting far more clicks per second out of a mouse than ordinary clicking, and this site publishes guides to them. We do not record which one a player used, and from a final score a server cannot tell a drag click from a script, so the desktop click-speed figure mixes techniques the way the combined figure mixed devices. We have not removed the fast scores on suspicion: we have no evidence that would distinguish them, and deleting a legitimate drag-clicker to tidy up a distribution would be the same kind of quiet dishonesty this page exists to avoid. It is one of the reasons every figure here is a median published with its sample, and never a maximum.

The charts are a separate matter, because a single 38-second "reaction" stretches an axis until nine players in ten share two bars. So each chart is drawn over Tukey's inner fences — the quartiles, extended by one and a half times the interquartile range, and never beyond the scores actually observed — which lets each test's own spread decide where its outliers begin instead of us naming a cut-off we cannot source. Scores beyond the fences are not squeezed into the end bars; they are left out of the drawing and counted underneath it, so every chart states how many players it does not show. The full first-to-ninety-ninth-percentile spread stays in the citation strings below and in the JSON, where nothing is trimmed at all.

Cite this data

These aggregated statistics are free to reuse under CC BY 4.0. Attribution means naming "GGReflex Global Reflex Statistics" and linking back to this page. Ready-made citation strings, the figure for the test you selected above, and the JSON endpoint are in the next panel. For research questions or data requests, see the contact page.

Frequently asked questions