Speed
Typing Speed Test
Type five sentences. Each one is timed from your first keystroke to your last, and your score is the median of the five in words per minute — with the share of characters you got right shown beside it rather than folded into it. A word here is five characters, which is a convention, and this page says so before it says anything else.
Typing speed
Five sentences, one at a time. Read each one, then type it as quickly and accurately as you can, and press Enter when you reach the end. The clock for each sentence starts on your first keystroke and stops on your last — so take as long as you like reading before you begin. Backspace works; pasting does not.
What your equipment contributed
- Your keyboard
- …inferred from pointer type, not detected
- Clock resolution
- …smallest step this browser reports
- Between your keystrokes
- not yet measuredmedian interval, this run
For once, the clock is not the problem. Both ends of the clock are keystrokes of yours, so a constant input lag delays the start and the stop equally and cancels out of the rate entirely — the same reason our aim trainer reports no display figure. The keyboard is the limit, and no web page can measure a keyboard. Key travel, actuation force, layout, how firmly the device is held, and on a phone the size of the keys and whatever autocorrect did before we saw the characters — all of it is in your number and none of it is separable from you. We do not invent a figure for any of it. We do report which kind of keyboard we think you used, because that decides which published range your result should be read against, and we could be wrong: a touchscreen laptop with a real keyboard attached will read as on-screen here.
How this is measured
What starts and stops the clock
Each sentence is its own trial. The clock starts on the first keystroke of that sentence and stops on the last, both timestamped with performance.now() as the first statement in the event handler. Reading time is therefore outside the measurement, and so is the pause before you press Enter. Pressing Enter submits and is never counted as a character.
The arithmetic, including the character we subtract
Words per minute is (characters − 1) ÷ 5 over the measured interval, in minutes. Five characters including spaces is the standard definition of a “word” in text-entry research and the reason any two WPM figures can be compared. The − 1 is there because the clock started on the first keystroke: no time at all was measured for that character, so counting it would credit you with work done before the clock existed. On a 56-character sentence that is worth a couple of WPM. The reference study subtracts it too.
How accuracy is scored, and why it stays separate
After you submit a sentence we compare what you typed with the target by edit distance — the fewest insertions, deletions and substitutions that turn one into the other — and report the share of the target's characters that survived. Edit distance rather than a position-by-position check, because dropping one character early shifts everything after it, and a positional comparison would score a single slip as a ruined sentence. Only uncorrected errors count: anything you fixed with backspace already cost you time, and time is where a correction belongs. We never multiply speed and accuracy into one figure — see the note under your result.
How the five sentences become one number
The reported figure is the median of the valid sentences, which is site policy for every timed test here (lib/trials.js): one stumbled sentence would drag a mean by several WPM while barely moving a median. Every sentence's own rate is listed under the result so the spread is visible rather than averaged away.
What gets discarded, and what you are told about it
Three things disqualify a sentence, and each is named on screen when it happens. A pause of more than 5 seconds between consecutive keystrokes means you stopped typing, and a pause corrupts that sentence's rate outright rather than nudging it. A rate above 300 WPM was not produced by fingers on keys — the fastest of the 168,960 people in the reference study averaged 158, and the fastest sustained human typing on record is a little over 200, so a sentence at 300 is a held key, a macro, or an autofill arriving as ordinary keystrokes we cannot tell apart. And a transcription below 75% correct was not a transcription of that sentence — a threshold we did not choose, but took from the reference study, which excluded participants whose error rate exceeded 25%. If fewer than three of your 5 sentences survive, no score is shown at all, because a median of two is not a measurement. Pasting is blocked outright.
Why there is no per-character feedback while you type
Most typing tests colour each letter green or red as you go. We deliberately don't. Chasing a live error highlight is a different task from transcribing a sentence — it changes when you look at the screen and when you decide to backspace — and we have no way to tell you how much of your number that difference is worth. Leaving it out costs a little polish and keeps the measurement one thing.
What your equipment contributes
Unusually for this site, almost nothing we can name. Both ends of the clock are keystrokes of yours, so a constant input lag delays the start and the stop by the same amount and cancels out of the rate — the same reasoning our aim trainer uses, and the reason lib/apparatus.js and its display figure are not used on this page at all. The panel above reports the one limit that is derivable: the resolution of your browser's clock, measured on your machine, which comes out a few thousand times smaller than the gap between two of your keystrokes. The keyboard is the real limit and no web page can measure one. Key travel, layout, key size, how the device is held, and on a phone whatever autocorrect changed before the characters reached us — all of it is in the number and none of it is separable. We invent no figure for any of it.
How your result compares
This is the first test on BlinkBench whose reference range was collected on the same kind of apparatus as yours — a web browser, on the participant's own hardware, wherever they happened to be sitting. Our other tests compare a screen against a laboratory; this one does not have to, and the comparison is correspondingly stronger.
| Keyboard | Participants | Mean WPM | Spread |
|---|---|---|---|
| Physical | 168,960 | 51.56 | SD 20.2 |
| On-screen | 37,370 | 36.2 | not published |
The physical-keyboard sample averaged 51.56 WPM (SD 20.2), with individual means spanning 4 to 158 WPM — a spread wide enough that the average is barely a landmark. Participants were young: mean age 24.5 (SD 11.2). Each typed 15 sentences drawn from a pool of 1,525. The mobile sample averaged 36.2 WPM with 2.3% uncorrected errors, and the authors report that people typing with one or two fingers reach roughly 70% of the speed they manage on a full keyboard.
What we matched, and what we could not
Matched: a browser-based transcription task on the participant's own hardware; the same WPM definition, including the subtracted first character; sentences built to the same constraints (≤ 70 characters, ≥ 3 words, simple punctuation only); the same 75% accuracy floor for counting an attempt at all.
Not matched: they used 15 sentences and we use 5, so our estimate is noisier than theirs — five trials is a usable median and not a stable one. They report a mean per participant; we report a median, because five trials and a mean is exactly the combination one bad sentence ruins. Their pools were drawn from email and newswire corpora; ours are fifteen sentences we wrote to the same rules, so a specific sentence you find awkward moves your number more than it would move theirs. And their participants opted into a typing test on a typing-test website, which is a self-selected sample in the same direction ours is.
Read 51.56 and 36.2 WPM as bearings on what typical looks like on each kind of keyboard — not as lines you have passed or failed.
No percentile is shown, even though these are much better matched samples than this site usually gets, and even though the desktop study publishes a mean and an SD. Typing speed is not normally distributed — it is right-skewed, with a long tail of trained typists — so placing you on a bell curve built from a mean and an SD would produce a confident number that is wrong in a direction we could not state. The sample also skews young (24.5 on average) and self-selected towards people who visit typing-test sites. And a five-sentence median is a noisier estimate than the fifteen-sentence mean the reference figure came from, so the precision a percentile implies does not exist. A percentile drawn from BlinkBench's own visitors would be worse on every one of those counts.
Physical keyboards: Dhakal, V., Feit, A. M., Kristensson, P. O., & Oulasvirta, A. (2018). Observations on Typing from 136 Million Keystrokes. Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems, 1–12.
On-screen keyboards: Palin, K., Feit, A. M., Kim, S., Kristensson, P. O., & Oulasvirta, A. (2019). How do People Type on Mobile Devices? Observations from a Study with 37,000 Volunteers. Proceedings of MobileHCI '19, 1–12.
Last reviewed: July 2026
Frequently asked questions
What counts as a “word” in words per minute?
Five characters, including spaces — not an English word. It is the convention every published text-entry study uses, including both studies cited on this page, and it is the only reason WPM figures from different sources can be compared at all. It also means the unit has a bias built in: a passage of long words scores lower than a passage of short ones for exactly the same finger movement. That is why our sentences are drawn from a pool built to the same constraints as the reference study's — at most 70 characters, at least three words, simple punctuation only.
Why is my accuracy shown separately instead of as a “net WPM”?
Because the formula that combines them is a convention someone chose, not a thing anyone measured. Different sites subtract a different penalty per error, so two of them can report different net scores from identical typing, and neither can tell you which of the two numbers moved. Speed and accuracy are both real measurements and they trade off against each other; we show you both and leave the trade-off where it belongs. This is the same restraint our aim trainer applies when it declines to normalise throughput for error rate.
Does the timer include the time I spend reading the sentence?
No. Each sentence's clock starts on your first keystroke and stops on your last, so you can read for as long as you like before you begin, and the gap between finishing and pressing Enter is not counted either. There is one consequence worth knowing: because the clock starts on the first keystroke, no time was measured for the first character, so the arithmetic counts one character fewer than you typed. The reference study makes the same subtraction. Without it, a short sentence would be credited a couple of WPM for work done before the clock existed.
Is my phone result comparable to my laptop result?
Not directly, and that is why this page cites two reference figures instead of one. A study of 37,370 people typing in a mobile browser averaged 36.2 WPM; a study of 168,960 people typing on physical keyboards averaged 51.56. The same person is a different instrument on each. The mobile study also found that people who type with one or two fingers reach only about 70% of the speed they manage on a full keyboard. We infer which kind of keyboard you used from your pointer type and say which figure applies — an inference we can get wrong, and say so.