At what age is your brain sharpest?

01 Key findings

1. Overall performance peaked at age 25. Averaged across every test, scores rose from 18 to 25 and then fell steadily; by 32 they were back to the level of 18-year-olds.iqmetrics.org analysis of Lumos Labs NCPT data, 756,948 adults (CC BY 4.0)
2. Each skill has its own peak. Response control peaked at 18 or younger, short-term memory at 20 to 23, processing speed and verbal reasoning at 25, pattern reasoning at 29 and math reasoning at 36.iqmetrics.org analysis of Lumos Labs NCPT data, 756,948 adults (CC BY 4.0)
3. Speed fades fastest. After its peak, processing speed fell about 0.32 standard deviations per decade, roughly 2.5 times as fast as pattern reasoning (0.13) or verbal reasoning (0.11).iqmetrics.org analysis of Lumos Labs NCPT data, 756,948 adults (CC BY 4.0)
4. The education gap holds at every age. Adults with a graduate degree in their 60s scored about the same as adults with a high-school education in their 50s (-0.31 vs -0.34).iqmetrics.org analysis of Lumos Labs NCPT data, 756,948 adults (CC BY 4.0)
5. Morning tests scored best. After adjusting for age, tests started at 8 am scored highest and tests started at 2 am lowest, a small gap of about 0.15 standard deviations.iqmetrics.org analysis of Lumos Labs NCPT data, 756,948 adults (CC BY 4.0)

02 Statistics to quote

03 At what age does each thinking skill peak?

Different thinking skills reach their best at different ages. Skills that depend on speed and on holding information in mind peak in the early twenties; reasoning with patterns and numbers peaks later. Earlier research found the same uneven rise and fall ^[3]; our explainer At what age does intelligence peak? walks through those studies. The bar shows the 95% range for each peak age, from 200 resamples of the data.

Figure 1

Peak age by thinking skill

Dot = peak age, bar = 95% range. *Peaked at 18, the youngest age in the data, so the true peak may be younger.

Response control*18 (18–18)

Short-term memory20 (18–21)

Visual search speed21 (21–22)

Divided attention22 (19–23)

Working memory22 (20–23)

Task switching24 (24–25)

Verbal reasoning25 (25–26)

Processing speed25 (25–25)

Pattern reasoning29 (29–30)

Math reasoning36 (35–37)

152025303540

Table 1

Every test, with its peak age and decline

Click a column heading to sort. Decline = average fall per decade after the peak, in standard deviations. † Taken by fewer than 50,000 people, so the range is wider.

Skill	Peak age	95% range	Decline per decade	People	Test
Response control	18	18	0.11	72,657	Go/no-go (v2)
Response control †	18	18	0.14	9,526	Go/no-go (v1)
Short-term memory	20	18-21	0.13	408,459	Forward memory span (v1)
Visual search speed	21	21-22	0.13	581,705	Trail making, part A (v1)
Working memory	22	20-23	0.14	314,496	Reverse memory span (v1)
Divided attention	22	19-23	0.15	172,098	Divided visual attention (v1)
Short-term memory	23	23-24	0.18	296,824	Forward memory span (v2)
Working memory	23	21-24	0.16	280,283	Reverse memory span (v2)
Task switching	24	24-25	0.1	578,034	Trail making, part B (v1)
Verbal reasoning	25	25-26	0.11	659,102	Grammatical reasoning (v1)
Processing speed	25	25	0.32	331,161	Digit symbol coding (v1)
Processing speed	25	25	0.31	310,938	Digit symbol coding (v2)
Balance reasoning †	27	18-30	0.25	8,153	Scale balance (v1)
Visual search †	28	22-33	0.17	8,370	Dual search (v1)
Pattern reasoning	29	29-30	0.13	638,535	Progressive matrices (v1)
Visual memory †	29	27-33	0.13	10,855	Object recognition (v1)
Complex memory †	30	25-39	0.1	8,881	Complex memory span (v1)
Math reasoning	36	35-37	0.14	746,010	Arithmetic reasoning (v1)
Attention cueing †	40	29-42	0.04	10,272	Posner cueing (v1)

04 How do thinking scores change with age?

The line shows the average score at each age across all tests, smoothed across neighbouring ages, in standard deviations from the average adult (0). It rises to age 25 and then falls steadily.

Figure 2

Average score across all tests, by age

Standard deviations from the average adult. Smoothed with a 3-year window.

All tests (average)

-1-0.500.5

-1.07

1835526986

Skills age differently. Processing speed starts highest and falls the furthest; math reasoning holds steady into the 40s; pattern reasoning falls the least.

Figure 3

Three skills compared, by age

Standard deviations from the average adult. Smoothed with a 3-year window.

Processing speedMath reasoningPattern reasoning

-1.5-1-0.500.5

-1.55-1.01-0.61

1835526986

05 Do other studies find the same peak ages?

The best-known study of peak ages, by Hartshorne and Germine (2015), combined 48,537 online volunteers with the norms of the Wechsler intelligence and memory tests and found the same staggered pattern: speed first, memory later, knowledge last ^[3]. Our data agrees on the order and differs on one number. Digit symbol coding, the standard processing-speed task, peaked at 25 here against about 18 to 19 there. Their online sample included teenagers; ours starts at 18, and the two groups were tested a decade apart.

Table 2

Peak ages in this study and in Hartshorne and Germine (2015)

Our peaks are from the NCPT (756,948 adults, 2013 to 2020); the 2015 figures are as that paper and its press coverage reported them.

Skill	This study (NCPT)	Hartshorne and Germine 2015	Note
Processing speed (digit symbol coding)	25	about 18 to 19	Same task type; our peak is later. Our youngest testers are 18.
Short-term memory (forward span)	20	mid 20s to early 30s (digit span)	Different versions of the span task.
Working memory (reverse span)	22	early to mid 30s (working memory for numbers)	Different versions of the span task.
Visual working memory	not in the NCPT	25
Pattern reasoning (matrices)	29	not reported
Math reasoning	36	not reported
Vocabulary	not in the NCPT (grammatical reasoning peaked at 25)	late 60s to early 70s online; 40s in older Wechsler norms	Word knowledge keeps growing; the NCPT measures reasoning, not vocabulary.
Emotion recognition from the eyes	not in the NCPT	40 to 60, a long plateau

06 Your age, in one sentence

Pick an age to see how the average adult of that age compared with the peak age for overall thinking performance and for five skills. Skills that peak later are still rising when others have started to fall. Every age has its own link and a card you can download or share.

07 Does more education slow the decline?

People with more education scored higher at every age, and every group followed the same downward curve. The gap between a graduate degree and a high-school education was about 0.40 standard deviations in the 30s and 40s and about 0.25 in the 70s and 80s.

Table 3

Average score by age and education

Average across all tests, in standard deviations from the average adult. HS = high school; Some college includes associate degrees.

Age	HS or less	Some college	Bachelor's	Graduate
18-29	0.16	0.36	0.51	0.52
30-39	0.02	0.21	0.38	0.42
40-49	-0.15	0.02	0.19	0.24
50-59	-0.34	-0.21	-0.07	-0.02
60-69	-0.6	-0.48	-0.35	-0.31
70-79	-0.89	-0.79	-0.68	-0.64
80-89	-1.22	-1.11	-1.01	-0.97

08 What time of day do people think best?

The dataset records the local hour each test was started. After removing the effect of age, scores were highest for tests started around 8 am and lowest around 2 am. The pattern is consistent across 712,071 complete tests, but the effect is small, and people who test at 2 am may differ from people who test at 8 am in ways the data cannot see.

Figure 4

Age-adjusted score by the hour the test was started

Difference from the daily average, in standard deviations.

All adults (age-adjusted)

-0.1-0.0500.05

-0.048

12 am5 am10 am3 pm8 pm11 pm

Men and women scored almost the same: after adjusting for age, the gap was about 0.05 standard deviations.

09 How we measured it

1. 01

   The data

   The NeuroCognitive Performance Test is a web-based test battery run by Lumos Labs. Its public dataset holds 5.45 million subtest scores from 757,427 adults, published under CC BY 4.0 ^{[1] [2]}.

2. 02

   Cleaning

   One test per person (no one took it twice), ages 18 to 89, one score per subtest. Scores were trimmed at the 0.5th and 99.5th percentiles and turned into standard scores within each subtest. For timed subtests, lower times count as better.

3. 03

   Peak age

   For each subtest we averaged scores at every age, smoothed the curve with a 3-year window, and took the highest point. Ages with fewer than 30 people were not used.

4. 04

   Uncertainty

   The 95% range comes from 200 resamples of the people who took each subtest.

5. 05

   Adjustments

   The overall score is the average of a person’s subtest scores. Time-of-day and sex comparisons remove the average effect of age first.

6. 06

   Checks

   The peak-finding method was tested on simulated data with known peaks of 25, 38 and 55 and recovered 24, 38 and 55. Two list-learning subtests taken by fewer than 1,600 people were left out.