A "5 min read" label is a text's length divided by an assumed reading speed. For English, the speeds that publishers, plugins and calculators print run from 200 to 300 words a minute, so 1,000 words comes out anywhere from 3 min 20 s to 5 min on our arithmetic. For Chinese the larger difference is what gets counted. We counted chapter 1 of Lu Xun's 阿Q正傳 three ways. It is 1,411 Chinese characters, or 1,023 words once software splits it into words, or 21 tokens if it is split at spaces. A tool that counts spaces can show a minute or less, while a rate per character gives several minutes.
Others noticed the Chinese mismatch first. Owen Young wrote in 2022 that his Zola blog's Chinese estimates ran far too long, and a Bricks Builder forum thread from January 2024 reported odd results for traditional Chinese. We applied these rates to the same texts for a direct comparison.
Silent readers in their own language average 238 words a minute
A meta-analysis pools many separate studies into one estimate. Marc Brysbaert's, in the Journal of Memory and Language in 2019, covers 190 studies with 18,573 participants. Its 238 words a minute (wpm) is the unweighted mean of silent-reading studies in languages written in the Latin alphabet. By our tally of its table, 144 of the 190 are in English, and those alone average 236.
The paper labels it non-fiction, limits it to adults aged 18 to 60 and calls it "only valid for native speakers". Many readers in this region read English as a second language, and the review gives no figure for them. Its 260 wpm for fiction comes from a separate study of 48 readers who each read one book.
Brysbaert also traced two commonly recommended rates, 250 and 300 wpm, and found his estimate 5% and 20% below them. Medium's help page bases its read time on "roughly 265 WPM", with an adjustment for images, and names no source.
Volunteers who read aloud underlie the rates in Yoast's table
The International Reading Speed Texts (IReST) are short passages in 17 languages for eye clinics. In the 2012 study that set their norms (abstract), 25 native readers per language, aged 18 to 35, read them aloud against a stopwatch. English came out at 228 wpm, Chinese at 255 characters a minute and Japanese at 357.
The WordPress plugin Yoast SEO says in its reading-time file that its numbers come from this research, and its table holds the same values. The two are therefore a single source, and both describe reading aloud. Yoast's 17.8 changelog records a fix for other languages being given English speeds, so those rates came later.
Brysbaert's own mean for reading aloud, from 77 studies, is 183 wpm. Aloud figures on the same Chinese passages differ as well: Han and colleagues timed 32 college students at 298 characters a minute in 2017, about 17% above IReST.
Book readers get through Austen faster and Darwin slower in Brysbaert's equation
The paper offers a correction "for a reading rate calculator for individual texts": 238 × 4.6 ÷ the text's mean word length in letters, 4.6 being the non-fiction average. We counted the Project Gutenberg texts of Pride and Prejudice at 4.43 letters a word and On the Origin of Species at 4.82. The equation gives Austen 247 wpm and Darwin 227, moving the two books in opposite directions from one default. It is not exact: on difficult texts from a 1978 study it predicted 203 wpm where readers managed 182.
Chinese readers meet tools that count different things
Chinese has no spaces between words, so a tool must decide what a word is. Word segmentation splits text into words by dictionary or rule. Our segmenter found 1,023 words in the chapter, 14 of them footnote numerals, and 1.26 to 1.48 characters a word across whole books. Brysbaert converted at 1.5, and the IReST passages work out at 1.61. His Chinese silent figure, 260 wpm from 18 studies, is about 390 characters a minute at 1.5, on our arithmetic.
Medium gives 500 characters a minute for "Chinese, Japanese and Korean". Zola changed its rule in version 0.23.0, released on 5 August 2026 (pull request 3145): it now counts Chinese characters at 439 a minute, citing a 2024 paper we could not open. Yoast's file, as published, maps a locale such as zh_CN to "zh" while its table lists Chinese under "cn". Read as written, Chinese would take the file's fallback rate on words split at spaces; we read the code and did not run it.
| Rule, as published | What it counts | Chapter 1 (our arithmetic) |
|---|---|---|
| Medium, 500 a minute | 1,411 characters, punctuation excluded (our choice) | 2 min 49 s |
| Zola since 0.23.0, 439 a minute | Characters | 3 min 13 s |
| Brysbaert silent, about 390 a minute | Characters | 3 min 37 s |
| IReST aloud, 255 a minute | Characters | 5 min 32 s |
| Zola before 0.23.0, 200 a minute | Each character as a word | 7 min 3 s |
| Yoast's code as we read it, 183.13 a minute | 21 tokens split at spaces | 1 min, rounded up |
Leaving out the aloud norms, the three current rules span 1.28 times on the chapter, narrower than English's 1.5. Brysbaert's table has no silent-reading row for Japanese, Malay or Thai. Whether a 字 counts as a word also comes up in Chinese typing tests.
Until 18 October 2026 our reading time calculator showed "0 sec" for Chinese, Japanese and Thai text beside a non-zero word count. It gave results to the second, with speed hints and a heading saying the label lifts engagement, none of which any source we found supports. The revision made that day times Chinese characters at IReST's 255 a minute and Japanese at 357, labelled as aloud rates, and rounds to the minute. Chapter 1 of Project Gutenberg text #25332 at the default rates now gives 6 min. Thai text is shown as not estimated.
Individual readers vary widely around the published average
Brysbaert writes that "most adults fall in the range of 175–300 wpm" for non-fiction, and 200 to 320 for fiction. That is about a quarter either side of 238, on our arithmetic, or 3 min 20 s to 5 min 43 s for 1,000 words. A range like this describes group performance in a study, not a benchmark for any single reader.
E-book app users read more slowly outside the laboratory
People in a study know they are timed, and Brysbaert expects reading "in less constrained situations" to be slower, through mind wandering and interruptions. In 2012 an e-reader company said the average reader finished Mockingjay, about 101,000 words, in seven hours, which he puts at 241 wpm, close to the laboratory mean.
A Russian e-book app's 8,000 users averaged 150 wpm over ten months. Russian needs about a fifth fewer words than English, he notes, so that is near 187 English words a minute. Those readers, choosing their own books in their own time, still came out below the 238 measured in studies.
View author profile → · Editorial policy
