How fast do humans think? 47 activities measured in bits per second

01 Key findings

1. Most of what people do runs at 1 to 40 bits per second. Our 47 activities have a median of 7.5 bits/s; 24 are below 10 and 20 of them within a factor of 2 of 10 (5 to 20).IQ Metrics dataset
2. Speech is the fast lane, at about 39 bits per second. Across 17 languages the fastest talkers pack the least into each syllable (r = -0.83), so information rates land close together.Coupe 2019; IQ Metrics calculation
3. Speed and stamina trade off. Memory champions reach 23.2 bits/s for 12 seconds of digits and 3.1 over an hour; Rubik’s Cube records run from 26 bits/s for a 2.51 s solve to 1.17 for an hour of blindfolded cubes.IAM, Memory League, WCA; IQ Metrics calculation
4. Brain chips decode 0.65 to 7.2 bits per second. The 2023 speech implants are the fastest, at a fifth of natural speech.Pandarinath 2017; Willett 2021, 2023; Metzger 2023; Card 2024
5. The 10 bits/s figure is contested. A 2025 Nature Neuroscience reply is titled “more than 10 bits per second”, and two 2025 preprints report up to 42 bits/s for memory and 215 for mental calculation.Sauerbrei 2025; Wiederhold 2025
6. The gap is about method. Speech at 39 bits/s counts information per syllable from a dictionary model; the 10 bits/s figure counts English at about 1 bit per character, the entropy a reader can predict.Shannon 1951; Coupe 2019; Zheng & Meister 2025

02 Statistics to quote

03 How fast do humans think?

Humans think at about 10 bits per second by the best-known estimate, and the 47 activities in our dataset have a median of 7.5 bits per second ^[1]. A bit is one yes-or-no answer: 10 bits a second is enough to pick one item out of about a thousand every second. Jieyu Zheng and Markus Meister at the California Institute of Technology (Caltech) reached the 10 bits/s figure by reviewing typing, speech, memory sports, speedcubing and video games, and they call it reliable “within a factor of 2 either way” ^[1].

Our dataset checks that claim one activity at a time. We took a named record or study for each activity and turned it into bits per second with one stated method: the source paper’s own figure where it computed one, characters per second times the entropy of English for text, or items per second times the bits in each item for digits, cards and cube states. 24 of 47 activities come out below 10 bits per second and 22 between 10 and 50. The one figure above 50 is a 2025 preprint on mental calculation (215 bits/s) ^[17].

Figure 1

Every activity on one scale: 24 of 47 run below 10 bits per second

Each dot is one activity, placed by its information rate (bits per second, log scale: each step is ten times faster), one row per type of activity (top to bottom: language, typing, memory sports, Rubik's Cube, video games, lab tasks, brain chips, mental calculation). The median is 7.5 bits/s; 20 of 47 sit between 5 and 20 bits/s, inside the factor of 2 that Zheng and Meister allow around their 10. Records and studies as of 4 November 2026.

Speaking, listening, readingTypingMemory sportsRubik's CubeVideo gamesLab tasks and perceptionBrain chipsMental calculation

Type of activity (one row each, see the colours)

Brain chip 2017Multi-blindDigits, 1 hourTypingHickSpeech chip 2023StarCraftSpeed cardsCube 3×3SpeechMental calculation

0.5125102050100200

Information rate, bits per second (log scale)

The spread is narrow for such different tasks. A brain-chip user copying sentences in 2017 managed 0.65 bits per second ^[8]; an average typist about 4.3 ^[5]; a memory champion looking at a deck of cards 17.7 ^[13]; a speaker in any of 17 languages about 39 ^[2]. Between the slowest and the fastest of the 47, the factor is about 330, set by one preprint; without it, the fastest is a reading task at 43 bits per second ^[7].

Figure 2

Most human activities run at 0 to 20 bits per second; one preprint claims 215

All 47 activities in bins of 10 bits per second. 24 are under 10 bits/s and 22 between 10 and 50; the long right tail is one preprint on mental calculation (215 bits/s, not yet peer reviewed). Dashed lines: Zheng and Meister's 10 bits/s and speech at 39 bits/s.

2413423110 bits/sSpeech 39

04080120160200

Bits per second

Bit

One yes-or-no answer. Picking one of 2 equally likely options takes 1 bit, one of 1,024 takes 10 bits. Information rate counts the bits a person resolves per second.

Bits per second (bits/s)

Information per unit of time. A typical Wi-Fi connection moves about 50 million ^[21]; human behaviour about 10 by the Zheng and Meister estimate ^[1].

Entropy

The average information in a symbol, given what came before. Shannon measured printed English at 0.6 to 1.3 bits per letter with 100 letters of context, “of the order of one bit” ^[6].

Information rate

Entropy per second: how much a stream of speech, text or moves narrows down what could have come next. A typist who copies predictable English at 10 keys a second transmits about 10 bits/s, not 10 times the 6 bits of a key ^[1].

Information density

Bits per syllable of a language, from a model of how predictable its syllables are; from 4.8 in Basque to 8.0 in Vietnamese ^[2].

Speed of thought

Not a measured quantity of its own. Researchers infer it from behaviour (what a person reads, says, types or remembers per second), which is what this page compares.

Sifting number

Zheng and Meister’s name for the ratio between what the senses take in (about a billion bits/s) and what behaviour uses (10): about 100 million ^[1].

Processing speed

In intelligence testing, how quickly a person does simple tasks such as matching symbols. Related to, but not the same as, information rate; see processing speed and IQ.

04 Where does "10 bits per second" come from?

The 10 bits per second figure comes from a 2025 review in the journal Neuron by Jieyu Zheng and Markus Meister, posted as a preprint in August 2024 and published online on 17 December 2024 ^[1]. Their first example is a typist: 120 words a minute is 10 keystrokes a second, and because English carries about 1 bit per character, the typist transmits about 10 bits per second. Their table lists 12 behaviours, from binary-digit memorisation (4.9 bits/s) to object recognition (30 to 50) and speech in 17 languages (39).

Two of the paper’s own numbers do not match its stated method, and we list both. For Tetris it prints “around 7 bits/s”, but its upper-bound method (4 orientations x 9 positions, 3 to 4 pieces a second) gives up to 20.7 bits/s (IQ Metrics recalculation); for StarCraft the text computes 16.7 bits/s and the table prints 10. The cube paragraph gives the number of cube states as 4.3 x 10^16; the correct count is 4.3 x 10^19, which is the “2^65” the paper uses, so its 11.8 bits/s stands. None of these changes the order of magnitude.

The figure spread fast. Caltech’s news release of 17 December 2024 said humans think “at a rate of 10 bits per second” and that the senses gather “a billion bits per second, which is 100 million times faster” ^[21]. The same release quotes the senior author describing “the trillion that our senses are taking in”, a thousand times the paper’s own figure. OpenAlex lists 58 works citing the Neuron paper by 4 November 2026 ^[22].

Figure 3

How "10 bits per second" and "a billion" travelled

Each step in time order; steps are evenly spaced. The behaviour figure travelled unchanged. The senses figure grew from a billion to "the trillion" in one quote.

Where the number beganRepeated, no new dataMass mediaFact-checkMeasured by a study

Thought runs at 10 bits per secondVerdict: a fair summary of the paper; contested

1. 2024Zheng & Meister preprint and Neuron paper: about 10 bits/s12 behaviours reviewed; "within a factor of 2 either way"
2. 2024Caltech news release: "a rate of 10 bits per second"17 December 2024
3. 2025Memory athletes: up to 42 bits/s in short tasksWiederhold, preprint
4. 2025Mental calculators: above 215 bits/sWiederhold, Stemmler and Herz, preprint
5. 2025Nature Neuroscience: "The brain works at more than 10 bits per second"Sauerbrei and Pruszynski (title read)
6. 2026IQ Metrics: 47 activities, median 7.5 bits/sThis page

The senses take in a billion bits per secondVerdict: an order of magnitude, quoted 1,000 times higher once

1. 2024Paper: about 10^9 bits/s; one eye's cones about 1.6 billionCapacity from 6 million cones x 270 bits/s
2. 2024News release: "a billion bits per second", "100 million times faster"Matches the paper
3. 2024Quote in the same release: "the trillion that our senses are taking in"1,000 times the paper's figure
4. 2025Calculation preprint: senses "exceeding 10^6 bit/s"A lower bound, different number
5. 2026Cognitive-reserve preprint: a "10^9-to-10 bits/s gap"Uses the paper's numbers as given

05 How fast does the brain process information from the senses?

The senses can carry about a billion bits per second, 100 million times more than behaviour uses, by Zheng and Meister’s estimate ^[1]. One cone photoreceptor in the eye can transmit about 270 bits per second; the 6 million cones of one eye about 1.6 billion. The optic nerve, a million fibres firing at up to 50 spikes a second at about 2 bits per spike, has room for about 100 million bits per second “or less”, so the retina already compresses the signal at least tenfold. A single neuron in the cortex transmits about 10 bits per second, the same as the whole person.

Figure 4

From 1.6 billion bits a second at the eye to 10 in behaviour

Orders of magnitude from Zheng and Meister: the cones of one eye can carry about 1.6 billion bits per second, the optic nerve about 100 million or less, and behaviour about 10. The last bar is drawn to scale, so it cannot be seen: it is one hundred-millionth of the first. These are capacities, not measurements of one person.

Cone photoreceptors of one eye6 million cones x about 270 bits/s each1,600,000,000all

The retina passes on about 1 bit in 16 (6.25%)

Optic nerve of one eye1 million axons x 50 spikes/s x 2 bits per spike, "or less"100,000,0006.2% of all

Behaviour keeps about 1 bit in 10 million of the optic nerve's capacity

Human behaviourthe 10 bits/s of Zheng & Meister100.0% of all

These are capacities worked out from cell physiology, not measurements of a person seeing, and the paper calls many of its numbers “rough estimates” good to a factor of 2. The point survives the uncertainty: the gap between input and behaviour is about eight orders of magnitude, which, they write, “may be the largest unexplained number in brain science”. The brain statistics fact check traces other famous brain numbers the same way.

06 How fast can humans talk? About 39 bits per second

Humans talk at about 39 bits per second, the mean information rate of 170 adults reading the same 15 texts in 17 languages (39.15 bits/s, SD 5.10) ^[2]. Christophe Coupe, Yoon Mi Oh, Dan Dediu and Francois Pellegrino measured speech rate in syllables per second (mean 6.63) and the information density of each language in bits per syllable, and multiplied the two. Their data are open under CC BY 4.0 ^[3], so we recomputed every language: our simple means give 39.2 bits/s overall, from 33.8 in Thai to 45.9 in French.

Figure 5

The fastest talkers pack the least into each syllable

Rank of 17 languages (1 = highest) on syllables per second, bits per syllable and bits per second. Japanese is spoken fastest (8.0 syllables a second) but ranks 16 of 17 on bits per syllable; Vietnamese packs the most per syllable and is spoken among the slowest. Rates: IQ Metrics means of the authors' data.

VietnameseJapaneseOther languages

Syllables per secondSyll/sBits per syllableBits/syllBits per secondBits/s

French 10English 11Vietnamese 16Spanish 2Mandarin Chinese 14Japanese 1Korean 7Finnish 4Serbian 6Catalan 8Italian 5Turkish 9German 12Basque 3Cantonese 15Hungarian 13Thai 17

1 French2 English3 Vietnamese4 Spanish5 Mandarin Chinese6 Japanese7 Korean8 Finnish9 Serbian10 Catalan11 Italian12 Turkish13 German14 Basque15 Cantonese16 Hungarian17 Thai

Languages trade speed for density. Japanese speakers produce 8.0 syllables a second, the most in the sample, but each syllable carries 5.0 bits; Vietnamese carries 8.0 bits per syllable at 5.3 syllables a second. Across the 17 language means, syllables per second and bits per syllable correlate at r = -0.83, and the products, bits per second, end up within about a third of each other ^[3].

Figure 6

Speed and density cancel out: r = -0.83 between syllables per second and bits per syllable

Pearson correlations across the 17 language means. Languages spoken faster carry less information per syllable, so their bits per second end up close together. Linked, not cause; 17 languages is a small sample.

-1 (opposite)+1 (move together)

Syll/sSRBits/syllIDBits/sIRBits per syllable-0.83Bits per second+0.27+0.29Syllable entropy-0.55+0.46-0.08

Figure 7

Seventeen languages, fifteen texts: every cell between 33 and 47 bits per second

Mean information rate in bits per second for each language (rows, fastest first) reading the same 15 short texts, shown in five groups of three texts (columns, in the authors' order), averaged over that language's speakers. Cells run from 33.2 to 46.8 bits/s. IQ Metrics calculation from the authors' open data (syllables / duration x bits per syllable); the CSV has the groups, the raw data every reading.

SlowerFaster

T1-31-3T4-64-6T7-97-9T10-1210-12T13-1513-15French46.546.346.145.045.8English46.345.442.943.346.8Vietnamese42.542.742.243.142.2Spanish41.841.341.743.641.3Mandarin Chinese40.741.240.141.040.9Japanese40.840.439.940.140.8Korean39.540.238.839.939.4Finnish40.141.338.638.538.4Serbian39.439.039.639.737.9Catalan38.138.738.539.239.4Italian36.239.340.837.937.1Turkish36.536.838.239.037.5German37.138.235.637.536.8Basque36.235.836.537.835.8Cantonese36.936.836.337.035.1Hungarian34.534.434.433.935.9Thai33.933.633.335.033.2

People differ more than languages. Speaker means run from 27.8 to 54.2 bits per second, while language means stay between 33.8 and 45.9; the paper also found lower means for women (by 1.01 bits/s) ^[2]. None of this ranks languages: the authors read 39 bits/s as a shared limit of the human communicative niche, and a 2026 study in PNAS found that understanding falls off above about 10 syllables a second ^[19].

Figure 8

Inside each language, speakers differ more than languages do

Slowest and fastest speaker in each language, each speaker's mean over the texts they read, in bits per second (fastest language first). Speakers run from 27.8 to 54.2 bits/s; language means only from 33.8 to 45.9. IQ Metrics calculation from the authors' data.

Slowest speakerFastest speaker

French40.252.6

English40.952.0

Vietnamese35.554.2

Spanish39.943.7

Mandarin Chinese35.946.3

Japanese36.544.6

Korean33.948.8

Finnish31.841.9

Serbian31.844.5

Catalan34.541.5

Italian29.546.7

Turkish32.346.2

German28.742.6

Basque31.741.4

Cantonese33.538.4

Hungarian27.840.2

Thai29.437.4

2030405060

Bits per second

Table 1

Speech in 17 languages: our means of the open data

Means over every reading of 15 texts (170 speakers, 2,288 readings). Bits per syllable is the authors' language-level measure. Our simple means differ slightly from the paper's model estimates (39.15 bits/s overall).

Language	Speakers	Syllables per second	Bits per syllable	Bits per second	Slowest to fastest speaker
French	10	6.88	6.68	45.9	40.2 to 52.6
English	10	6.34	7.09	44.9	40.9 to 52.0
Vietnamese	10	5.30	8.02	42.5	35.5 to 54.2
Spanish	10	7.73	5.43	42.0	39.9 to 43.7
Mandarin Chinese	10	5.86	6.96	40.8	35.9 to 46.3
Japanese	10	8.04	5.03	40.4	36.5 to 44.6
Korean	10	7.12	5.56	39.6	33.9 to 48.8
Finnish	10	7.17	5.49	39.4	31.8 to 41.9
Serbian	10	7.15	5.47	39.1	31.8 to 44.5
Catalan	10	7.07	5.49	38.8	34.5 to 41.5
Italian	10	7.16	5.29	37.9	29.5 to 46.7
Turkish	10	7.05	5.34	37.6	32.3 to 46.2
German	10	6.09	6.08	37.0	28.7 to 42.6
Basque	10	7.54	4.83	36.4	31.7 to 41.4
Cantonese	10	5.57	6.53	36.4	33.5 to 38.4
Hungarian	10	5.87	5.90	34.6	27.8 to 40.2
Thai	10	4.70	7.19	33.8	29.4 to 37.4

07 How fast can humans read, in bits per second?

Adults read English non-fiction silently at 238 words per minute, about 22.2 bits per second at 1 bit per character, by Marc Brysbaert’s 2019 meta-analysis of 190 studies with 18,573 readers ^[4]. Fiction goes faster, 260 words per minute (22.5 bits/s), because its words are shorter: 4.2 letters against 4.6. Reading aloud runs at 183 words per minute (17.1 bits/s). The range from 0.6 to 1.3 bits per character, Shannon’s bounds, moves silent non-fiction reading from 13.3 to 28.9 bits/s ^[6].

The record for reading is older. In 1957 John Pierce and John Karlin at Bell Labs had people read word lists aloud as fast as they could and found about 43 bits per second for rapid readers, a lower bound “close to a limiting rate”; an earlier MIT report had found 30 to 35 bits/s for random dictionary words ^[7]. Zheng and Meister, working from eye fixations of 7 to 9 letters every 200 to 250 milliseconds, give 28 to 45 bits per second ^[1]. Brysbaert cites a 2016 review showing that more than 1,000 words a minute is impossible without severe loss of understanding. For what reading does for test scores, see reading and IQ.

08 How fast can humans type, in bits per second?

The average typist enters 51.56 words per minute, about 4.30 bits per second, in the largest typing study: 168,960 volunteers who copied 15 sentences each, 136 million keystrokes in all ^[5]. A “word” there is any 5 characters, so 51.56 words per minute is 4.3 characters a second, and at 1 bit per character, 4.30 bits per second (2.6 to 5.6 across Shannon’s range). The fastest tenth averaged 89.56 words per minute (7.5 bits/s), the slowest tenth 20.91 (1.7), and the fastest typists passed 120 words per minute, the 10 bits/s of Zheng and Meister’s example ^[1].

Typing is slower than speech because fingers are slower than the voice, but the gap is smaller than it looks in characters. Zheng and Meister’s point is that a professional typist and a gamer at 1,000 actions a minute (16.7 bits/s on their reckoning) land in the same range. To test your own speed on a reaction task, see the reaction time test and average reaction time by age.

09 How fast can memory champions take in information?

The fastest memory records run at 13 to 23 bits per second for 11 to 17 seconds of memorising, and fall to 2 to 3 bits per second over an hour. The speed-cards record, one shuffled deck memorised in 12.74 seconds, is 17.7 bits per second, because the order of a deck holds 225.6 bits ^[13]; Zheng and Meister print the same 17.7 ^[1]. The online Memory League record of 80 digits in 11.45 seconds is 23.2 bits/s ^[14]. Over an hour the numbers record (3,412 digits) is 3.1 bits/s and the cards record (1,829 cards) 2.2.

Figure 9

Memory champions take in 23.2 bits a second for 12 seconds, 3.1 over an hour

World records turned into bits per second of memorising time (a digit = 3.32 bits, a card in a shuffled deck = 4.34, a binary digit = 1). First column: 80 digits in 11.45 s (Memory League, online) and one deck in 12.74 s (IAM); second column: digits 15 minutes, cards 10 minutes. The 5-minute and binary-digit records are in the table below. Records as of 4 November 2026.

Decimal digitsPlaying cards

About 12 s12 s10-15 min10-15m30 min30 m1 hour1 h

Decimal digits 23.2Playing cards 17.7

3.1 Decimal digits2.2 Playing cards

A 2025 preprint by Bastian Wiederhold reads the same records and reports “up to 42 bit/s” in short tasks, with rates that fall with task length as a power law, and less than 1 bit/s for untrained people learning numbers ^[16]. Our figures count memorising time only, not recall, and use the score as printed. All records come from our memory world records study, which reads the International Association of Memory (IAM) and Memory League directly. To try a short version yourself, see the digit span test or the sequence memory game, and read what the research says about photographic memory.

10 How many bits per second is solving a Rubik's Cube?

A Rubik’s Cube holds 65.23 bits, the log of its 43,252,003,274,489,856,000 states, so the 2.51-second world record for one solve is at most 26.0 bits per second ^[15]. “At most”, because competitors may inspect the cube for up to 15 seconds before the timer starts (World Cube Association Regulation A3a1); counted with that time, the rate is 3.7. The average-of-five record is 18.6 bits/s, one-handed 11.6 and with the feet 4.2.

Blindfolded solving separates seeing from doing. The 11.56-second blindfolded record is 5.6 bits/s over the whole attempt; if, as Zheng and Meister report, about half goes on looking, the looking phase runs at about 11.3 ^{[15] [1]}. Their own example, the 12.78-second record of 2023 with 5.5 seconds of inspection, gives 11.8 bits/s. The multi-blind record, 63 of 65 cubes memorised and solved in 58 minutes 23 seconds, works out at 1.17 bits per second. For what cubing does and does not do for intelligence, see whether the Rubik’s Cube raises IQ.

Figure 10

Fast and simple or slow and rich: the same 1 to 40 bits per second

39 activities whose items can be counted, by items per second and bits per item, both on log scales; bubble area = bits per second. A Rubik's Cube state carries 65.23 bits but takes seconds; a typed character carries about 1 bit and comes 4 times a second. Products land in the same narrow band.

Speaking, listening, readingTypingMemory sportsRubik's CubeVideo gamesLab tasks and perceptionBrain chipsMental calculation

Bits per second11040

Bits per item (log scale)

125102050

ObjectsSpeechCube singleCube averageSpeed cardsStarCraft80 digits, liveOne-handedFast typistsTetrisTypingDigits 1 hCards 30 minCards 1 h

0.010.1110

Items per second (log scale)

Different activities reach the same band by different routes. A cube state is rich (65.2 bits) but comes once every few seconds; a typed character is poor (about 1 bit) but comes several times a second; a syllable sits in between at about 6 bits. Multiply items per second by bits per item and most activities land between 1 and 40 bits per second.

11 Can a brain chip make you think or communicate faster?

No brain chip yet beats natural speech: the fastest published speech implant decodes 78 words per minute, about 7.2 bits per second, a fifth of the 39 bits/s of speech ^[11]. A Stanford speech implant reached 62 words per minute (5.7 bits/s) ^[10], and a UC Davis implant has run for 248 hours of everyday conversation at about 32 words per minute (2.9) ^[12]. These figures convert words per minute at 5.5 characters a word (Shannon’s 4.5 letters plus a space) and 1 bit per character.

Earlier chips were slower. In 2017 a participant moving a cursor reached 3.7 bits per second on a grid of random targets, but copying English sentences at 39.2 correct characters a minute is only 0.65 bits/s, because English is predictable ^[8]. Imagined handwriting reached 90 characters a minute in 2021, 1.5 bits/s ^{[9] [1]}. These implants restore communication to people who have lost it; they are not a shortcut past the 10 bits/s of behaviour. Our count of people with brain implants lists every programme.

Figure 11

The best brain chip decodes 7.2 bits a second; speech carries 39

Best published rates of brain chips for people with paralysis, against people speaking, listening and typing. Words per minute converted at 5.5 characters a word and 1 bit per character (IQ Metrics calculation); the 2017 grid and 2021 handwriting rates are the papers' or Zheng and Meister's figures.

People: speaking, 17 languages39.1 bits/s

People: listening to narration13.0 bits/s

People: typing, average4.3 bits/s

Chip 2023: speech, surface grid7.2 bits/s

Chip 2023: speech, Utah arrays5.7 bits/s

Chip 2017: random grid3.7 bits/s

Chip 2024: everyday speech2.9 bits/s

Chip 2021: handwriting1.5 bits/s

Chip 2017: copying sentences0.7 bits/s

12 Is "10 bits per second" right? The dispute

The 10 bits per second figure is contested, and the main gap is about method, not data. Speech at 39 bits per second counts the information in each syllable from a model of the language ^[2]; the 10 bits/s figure counts English text at about 1 bit per character, the entropy left after a reader predicts what comes next ^{[6] [1]}. Zheng and Meister point out that the 39 bits/s leaves out much of the redundancy of language, as the speech paper’s methods acknowledge. Counted the same way as typing, at 1 bit per character, listening at 160 words a minute is 13 bits/s and silent reading 22.2.

Other researchers argue for higher numbers. A 2025 Nature Neuroscience piece by Britton Sauerbrei and J. Andrew Pruszynski is titled “The brain works at more than 10 bits per second” ^[18]; we could read only its title, venue and pages. Two 2025 preprints from the same first author report up to 42 bits/s for memory athletes and above 215 bits/s for mental calculators, split into 125 bits/s of perception and 90 of calculation ^{[16] [17]}. Both are preprints and not yet peer reviewed.

Figure 12

Published estimates of human information rate, 1952 to 2025

Each estimate with its range where the authors give one. Not shown on this axis: the 2025 mental-calculation preprint (215 bits/s), the senses (about a billion) and the Nature Neuroscience reply titled "more than 10 bits per second", which we could not read. As printed in each source; ranges are the authors'.

Hick 1952: choice reaction5.0 (5.0–5.0)

Fitts 1954: aimed movements11.0 (10.0–12.0)

Licklider 1954: random words aloud32.5 (30.0–35.0)

Sziklai 1956: naming objects40.0 (30.0–50.0)

Pierce & Karlin 1957: rapid reading43.0 (40.0–45.0)

Pandarinath 2017: brain chip3.7 (1.4–3.7)

Coupe 2019: speech39.1 (34.0–44.2)

Willett 2021: handwriting chip1.5 (1.5–1.5)

Zheng & Meister 2025: behaviour10.0 (5.0–20.0)

Wiederhold 2025: memory athletes42.0 (42.0–42.0)

01020304050

Figure 13

Seventy-five years of measuring the speed of thought

Each dot is a published estimate or a record, by date; hover for the number. Dates within a year are approximate where the source gives only the year or month. As of 4 november 2026.

PaperTechnical reportPreprintPress release

Language and reading5 estimatesShannon: 1 bit per letterCoupe: 39 bits/s

Movement and perception3 estimatesHick: about 5 bits/s

Brain chipsbest ratesPandarinath: 3.7 bits/s

The 10 bits/s debatepaper, press, repliesZheng & Meister preprint

19501960197019801990200020102020

Our data support the order of magnitude and not the exact number: the median is 7.5 bits/s, 20 of 47 activities fall within a factor of 2 of 10, and the long-standing estimates cluster between 5 and 45. The more an activity is a sprint over seconds, the higher its rate; the longer it runs, the closer it falls to a few bits per second. Our peak brain age study shows how processing speed changes across a lifetime.

13 Does thinking faster mean a higher IQ?

Not directly: information rate is a property of a task, and none of the 47 activities here is an intelligence test, so no rate on this page converts to an IQ. IQ tests do include processing-speed subtests, and faster performance on them goes with higher scores on average, as processing speed and IQ explains; memory span and reasoning are related but separate (working memory and reasoning, memory and IQ).

14 Every activity and estimate in our dataset

Table 2

Every activity in our dataset (47 rows)

Sorted by bits per second. Low to high = the entropy range (0.6 to 1.3 bits per character) for text rows, or the range the source gives. The CSV adds items per second, bits per item, the calculation and record holders.

Activity	Type	Human figure	Year	Bits per second	Low to high	Method	Source
Brain chip: copying English sentences by point-and-click (Stanford, 2017)	Brain chip	39.2 correct characters per minute	2017	0.65	0.4 to 0.8	Characters x bits per character	Pandarinath 2017
Multiple cubes blindfolded (WCA record)	Puzzle	63 of 65 cubes in 58:23	2025	1.17	–	Items x bits per item	WCA records
Brain chip: imagined handwriting (Stanford, 2021)	Brain chip	90 characters per minute	2021	1.50	–	Paper’s own figure	Willett 2021
Typing, slowest 10% of volunteers	Typing	20.91 words per minute (group mean)	2018	1.74	1.0 to 2.3	Characters x bits per character	Dhakal 2018
Playing cards, 1 hour (IAM record)	Memory sport	1,829 cards in 60 min	2019	2.20	–	Items x bits per item	IAM records
Brain chip: everyday speech over 248 hours (UC Davis, 2024)	Brain chip	32 words per minute	2024	2.93	1.8 to 3.8	Characters x bits per character	Card 2024
Playing cards, 30 minutes (IAM record)	Memory sport	1,270 cards in 30 min	2026	3.06	–	Items x bits per item	IAM records
Random digits, 1 hour (IAM record)	Memory sport	3,412 digits in 60 min	2023	3.15	–	Items x bits per item	IAM records
Spoken numbers, read aloud at 1 digit a second (IAM record)	Memory sport	456 digits in a row	2015	3.32	–	Pace x bits per item	IAM records
Binary digits, 30 minutes (IAM record)	Memory sport	6,270 binary digits in 30 min	2017	3.48	–	Items x bits per item	IAM records
Brain chip: point-and-click on a 6×6 grid of random targets (Stanford, 2017)	Brain chip	3.7 bits per second (best of 3 participants; others 2.2 and 1.4)	2017	3.70	1.4 to 3.7	Paper’s own figure	Pandarinath 2017
Random digits, 30 minutes (IAM record)	Memory sport	2,132 digits in 30 min	2026	3.94	–	Items x bits per item	IAM records
Playing cards, 10 minutes (IAM record)	Memory sport	550 cards in 10 min	2019	3.98	–	Items x bits per item	IAM records
Rubik’s Cube 3×3 with feet (WCA record)	Puzzle	15.56 s	2019	4.19	2.1 to 4.2	Items x bits per item	WCA records
Typing, average of 168,960 volunteers	Typing	51.56 words per minute (SD 20.2)	2018	4.30	2.6 to 5.6	Characters x bits per character	Dhakal 2018
Random digits, 15 minutes (IAM record)	Memory sport	1,300 digits in 15 min	2019	4.80	–	Items x bits per item	IAM records
Binary digits, 5 minutes (IAM record)	Memory sport	1,467 binary digits in 5 min	2019	4.89	–	Items x bits per item	IAM records
Choice reaction: press the key for the lamp that lit (Hick 1952)	Lab task	2 to 10 lamps, reaction times of hundreds of ms	1952	5.00	–	Paper’s own figure	Zheng & Meister 2025
Rubik’s Cube 3×3 blindfolded, whole attempt (WCA record)	Puzzle	11.56 s, look and solve	2026	5.64	5.6 to 11.3	Items x bits per item	WCA records
Brain chip: attempted speech, Utah arrays (Stanford, 2023)	Brain chip	62 words per minute	2023	5.68	3.4 to 7.4	Characters x bits per character	Willett 2023
Random digits, 5 minutes (IAM record)	Memory sport	630 digits in 5 min	2021	6.98	–	Items x bits per item	IAM records
Tetris, top-tier players (3-4 pieces a second)	Video game	3 to 4 pieces per second	2024	7.00	7.0 to 20.7	Paper’s own figure	Zheng & Meister 2025
Brain chip: attempted speech, surface grid (UCSF, 2023)	Brain chip	78 words per minute	2023	7.15	4.3 to 9.3	Characters x bits per character	Metzger 2023
Typing, fastest 10% of volunteers	Typing	89.56 words per minute (group mean)	2018	7.46	4.5 to 9.7	Characters x bits per character	Dhakal 2018
Typing, advanced typist (120 words per minute)	Typing	120 words per minute	2024	10.00	–	Paper’s own figure	Zheng & Meister 2025
Thinking of a thing (the Twenty Questions game)	Perception	20 bits in a few seconds	2024	10.00	–	Paper’s own figure	Zheng & Meister 2025
Aimed arm movements between targets (Fitts 1954)	Lab task	10 to 12 bits/s in 20-second sprees	1954	11.00	10.0 to 12.0	Paper’s own figure	Zheng & Meister 2025
Rubik’s Cube 3×3 one-handed (WCA record)	Puzzle	5.62 s	2026	11.61	3.2 to 11.6	Items x bits per item	WCA records
Rubik’s Cube blindfolded, looking phase only (12.78 s record of 2023)	Puzzle	65 bits in about 5.5 s of inspection	2023	11.80	–	Paper’s own figure	Zheng & Meister 2025
Listening to narration, English, 160 words per minute	Language	160 words per minute	2024	13.00	–	Paper’s own figure	Zheng & Meister 2025
Memory League cards, live (52 cards)	Memory sport	16.86 s for 52 cards	2016	13.38	–	Items x bits per item	Memory League
Memory League numbers, live (80 digits)	Memory sport	16.85 s for 80 digits	2018	15.77	–	Items x bits per item	Memory League
StarCraft, professional at 1,000 actions per minute	Video game	1,000 actions per minute	2024	16.70	10.0 to 16.7	Paper’s own figure	Zheng & Meister 2025
Reading aloud, English (adults)	Language	183 words per minute (77 studies, 5,965 readers)	2019	17.08	10.2 to 22.2	Characters x bits per character	Brysbaert 2019
Speed cards: one shuffled deck (IAM record)	Memory sport	12.74 s for 52 cards	2018	17.71	–	Items x bits per item	IAM records
Memory League cards, online (52 cards)	Memory sport	12.25 s for 52 cards	2025	18.41	–	Items x bits per item	Memory League
Rubik’s Cube 3×3, average of 5 (WCA record)	Puzzle	3.51 s	2026	18.58	3.5 to 18.6	Items x bits per item	WCA records
Silent reading, English non-fiction (adults)	Language	238 words per minute (190 studies, 18,573 readers)	2019	22.21	13.3 to 28.9	Characters x bits per character	Brysbaert 2019
Silent reading, English fiction (adults)	Language	260 words per minute	2019	22.53	13.5 to 29.3	Characters x bits per character	Brysbaert 2019
Memory League numbers, online (80 digits)	Memory sport	11.45 s for 80 digits	2024	23.21	–	Items x bits per item	Memory League
Rubik’s Cube 3×3, single solve (WCA record)	Puzzle	2.51 s	2026	25.99	3.7 to 26.0	Items x bits per item	WCA records
Reading, from eye fixations (7-9 letters per 200-250 ms)	Language	28 to 45 characters per second	2024	36.50	28.0 to 45.0	Paper’s own figure	Zheng & Meister 2025
Speaking, average of 17 languages	Language	39.15 bits/s (SD 5.10) from 170 speakers reading 15 texts	2019	39.15	34.0 to 44.2	Paper’s own figure	Coupe 2019
Naming objects in a 0.5-second glimpse (Sziklai 1956)	Perception	1 to 3 objects in 500 ms	1956	40.00	30.0 to 50.0	Paper’s own figure	Zheng & Meister 2025
Memory athletes, shortest tasks (Wiederhold 2025, preprint)	Memory sport	up to 42 bits/s	2025	42.00	–	Paper’s own figure	Wiederhold 2025 (preprint)
Reading lists of words aloud as fast as possible	Language	about 43 bits/s for rapid readers (a lower bound)	1957	43.00	40.0 to 45.0	Paper’s own figure	Pierce & Karlin 1957
Mental calculation, shortest competition tasks (Wiederhold 2025, preprint)	Mental calculation	above 215 bits/s (125 perceiving + 90 calculating)	2025	215.00	125.0 to 215.0	Paper’s own figure	Wiederhold 2025 (preprint)

Table 3

Every published estimate we found (17)

"Via" = we read the figure in a later paper that reports it, not in the original. Abstract = the paper's abstract only.

Year	Who	What	Estimate	Range	Unit	How we read it
1951	Shannon	Entropy of printed English, 100 letters of context	1	0.6 to 1.3	bits per letter	full text
1952	Hick	Choice reaction time: rate of gain of information	5	–	bits/s	via Zheng & Meister
1954	Fitts	Aimed movements of the arm	11	10 to 12	bits/s	via Zheng & Meister
1954	Licklider, Stevens and Hayes	Reading random dictionary words aloud	32.5	30 to 35	bits/s	via Pierce & Karlin
1956	Sziklai	Naming objects in a 0.5 s glimpse	40	30 to 50	bits/s	via Zheng & Meister
1957	Pierce and Karlin	Reading chosen word lists aloud, rapid readers	43	40 to 45	bits/s	full text
2017	Pandarinath et al.	Brain chip: point and click on a grid	3.7	1.4 to 3.7	bits/s	full text
2019	Coupe, Oh, Dediu and Pellegrino	Speech in 17 languages	39.15	34.05 to 44.25	bits/s	full text + data
2021	Willett et al. (as computed by Zheng & Meister)	Brain chip: imagined handwriting	1.5	–	bits/s	via Zheng & Meister
2025	Zheng and Meister	Human behaviour in general (“10 bits/s”)	10	5 to 20	bits/s	full text (arXiv v2)
2025	Zheng and Meister	Sensory input, all senses	1 billion	–	bits/s	full text (arXiv v2)
2024	Caltech news release	Thought “10 bits per second”; senses “a billion bits per second”	10	–	bits/s	full text
2025	Wiederhold (preprint)	Memory athletes, shortest tasks	42	–	bits/s	abstract
2025	Wiederhold, Stemmler and Herz (preprint)	Mental calculation, shortest tasks	215	125 to 215	bits/s	abstract
2025	Sauerbrei and Pruszynski	“The brain works at more than 10 bits per second” (Nature Neuroscience)	–	–	bits/s	title and venue
2026	Gao et al.	Speech comprehension falls above about 10 syllables a second	10	–	syllables/s	abstract
2026	Yin (preprint)	Builds a theory of cognitive reserve on the 10 bits/s gap	10	–	bits/s	abstract

15 How we built the dataset

1. 01

   Pick a record or study per activity

   One named source per row: a journal paper, a records body (International Association of Memory, Memory League, World Cube Association), or our own earlier studies of memory records and brain implants, which cite their primary sources. Blogs and news were leads only. Guinness World Records data were not used.

2. 02

   Read every number by script

   Every source was saved and every figure printed here was matched by a script against the saved text (57 checks); a figure that is not found stops the build. Paywalled papers are cited for facts only.

3. 03

   One method per row

   The paper’s own bits per second where it computed them; otherwise characters per second x bits per character for text (Shannon: 0.6 to 1.3, central 1.0), or items per second x the exact bits per item (digit 3.322, card in a deck 4.338, deck 225.6, cube state 65.23). Words per minute: 5 characters a word for typing, 5.6 or 5.2 for reading (Brysbaert’s word lengths plus a space), 5.5 for speech.

4. 04

   Keep low and high

   Where the entropy assumption matters, the row carries a low and a high. Sighted cube solves carry a low that adds the 15 seconds of untimed inspection.

5. 05

   Recompute what is open

   The speech data of Coupe et al. are open under CC BY 4.0, so we recomputed every language from 2,288 readings; the paper’s model figures are quoted next to ours.