Do yourself a favor. Stop typing.

Latch works in any Windows app with a text cursor: Outlook or Gmail in the browser, Slack, Teams, Word, Notion, VS Code, a terminal. It pastes where the cursor already is, so there is nothing to integrate. It drops the ums, false starts, and self-corrections, fixes obvious slips, and keeps your meaning, names, and tone. Latch is Windows only, and the interface is English.

Pauses, corrections, and the time to paste are real, and Latch has not been measured for speed or accuracy. This is a public preview: it is free, there is no payment, you get 60 recorded minutes a month, a reminder after that, and a take is never cut off.

• Typing default, 40. 40 is a round default next to a median of 39 net words per minute. It is not a measurement of the person using the calculator, and it is not a Latch result.
• TypingTest.now. 100,079 valid tests on TypingTest.now from 11 April to 6 July 2026: median 39 net words per minute, mean 42.5. Net words per minute count five characters as one word and subtract an error penalty. The tests are from people who chose to take one, at mixed lengths. They are not a sample of all adults, and they are not a study of a desktop keyboard at work. On the same page the 90th percentile is 69 and the 95th is 83, and a category label (not a measured average) calls a trained touch typist 65 to 80. At those speeds the gap against speaking is smaller.
• Ratatype. Ratatype, 30 April 2026, 506,024 tests: the median sits in the 30 to 40 words-per-minute band. That page does not publish a single mean of 41.4. A separate Ratatype infographic says the average is 41.4 and does not give this sample size, so the two figures are not combined here. The infographic is the other page.
• Speaking default, 150, National Center for Voice and Speech. The National Center for Voice and Speech says, in a tutorial aside, that the average rate of speech for English speakers in the United States is about 150 words per minute. The page gives no sample size and no method. It is a rule of thumb, not a corpus measurement, and it is not a measurement of dictation into an app.
• Pauses allowance, 100. 100 words per minute is an allowance this page chooses so pauses and corrections have room. It is not a measured speaking rate and not a Latch result.
• Speech against a phone keyboard, Ruan, Wobbrock, Liou, Ng, and Landay. The published paper (Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 2017; an earlier version circulated in 2016) reports English entry at 153 words per minute by speech and 52 on the keyboard, 2.93 times, for short-message transcription. The keyboard was the iPhone 6 Plus touchscreen, in a quiet lab, with people seated. The published study has 24 English speakers among 48 participants, ages 19 to 32. The 2016 preprint of the same work used 16 English speakers among 32 and reported different English means (about 161 and 53). Speech had a lower corrected error rate during entry and a slightly higher uncorrected error rate in the finished text (the 2017 abstract: 5.30% versus 11.22% corrected, 1.30% versus 0.79% uncorrected). It is not a desktop keyboard, not composition of the writer’s own text, and not Latch. The 2016 preprint is the earlier version.
• The typewriter, Smithsonian, Sholes & Glidden typewriter (ME.276068). The Smithsonian record for its Sholes & Glidden typewriter (ME.276068) says Remington put the machines on the market in 1874. That is the date of the machine as sold. It is not the date of any keyboard layout, and it says nothing about how fast anyone types.
• Words a day, 3,000. An example this page chooses. It is not a typical Latch day.
• The year. A year here is the number of days in the field, 365 unless you change it. It is not a work calendar. Hours are minutes times that many days, divided by 60.

Last updated October 18, 2026.