The Verification Venue · a thing everyone half-believes
The Thought That Barely Costs You
Your brain is an energy hog: about 20% of everything you burn at rest, on just 2% of your body mass. So surely a hard day of thinking burns through it? Barely. That bill is nearly fixed whether you daydream or do calculus. Hard thinking adds only about 5%: a snack's worth over a whole workday.
Do you burn more calories thinking hard? Barely.
Here is the strange part, established by brain-imaging work: the brain's enormous energy use is almost constant. Roughly 60–80% of it just keeps neurons talking to each other, all day, every day, awake or asleep. The extra a hard mental task demands is, in the words of the researchers who measured it, “remarkably small, often <5%” of that baseline — and across the whole brain, as little as 0.5–1%. Move the meter and watch how little that is.
The one slider that matters most: heavier body → higher resting metabolism → bigger brain-energy bill.
From idle daydream to the hardest exam of your life. We apply the full 5% bump to your entire brain — deliberately generous to the “thinking burns calories” idea.
A full, brutal working day of concentration.
Your resting metabolism
1,618 kcal/day
Mifflin–St Jeor equation
Brain's baseline (20%)
324 kcal/day
= 16 W, running whatever you do
Extra fuel your brain burns from thinking that hard, for that long
+5.4 kcal
over 8 h — that’s about 1 almond
Notice what the meter refuses to do: it will not give you a number big enough to matter. Crank the body weight to 140 kg, the effort to maximum, the hours to sixteen — the extra still lands shy of a banana. You cannot meaningfully lose weight by thinking, because the expensive part of your brain was already running before you started.
Effort vs. energy — and the curve that inverts
Plot energy against effort and the human brain gives you a nearly flat line: the baseline towers over anything a task adds. Now flip the meter to an AI language model — the one answering questions like this — and the same axes give you the opposite shape. This is the signature of the page, and it is honest only if stated carefully.
Human brain: the line is almost flat. Baseline (~80% of the budget) dominates; the task-evoked bump is the thin sliver on top. Working harder does not meaningfully raise the total.
The check — every number recomputed in front of you
These lines recompute from your slider values as you move them. Nothing here is stored; it is arithmetic on published constants.
Every free choice and uncertainty, named:
- The 20% brain share is the classic figure (brain = 2% of mass, ~20% of resting energy; Scientific American; Raichle & Mintun 2006). Real values run ~16–25% by person and method.
- The 5% bump is Raichle & Mintun’s upper figure for local task-evoked increase. We apply it to the whole brain — a deliberate overestimate. Whole-brain increase is really nearer 0.5–1%, so the true extra is smaller than the meter shows.
- The RMR formula is Mifflin–St Jeor, the modern clinical standard; it is an estimate (±~10%) and needs height, age and sex, defaulted here and adjustable above.
- A banana is taken as 105 kcal (USDA, medium banana). The comparison is illustrative.
- The AI side below is direction only. More reasoning tokens cost more energy, monotonically — but the per-query joules are not public and estimates vary by orders of magnitude, so no number is drawn to scale and none is claimed.
Run it offline: node research/does-thinking-hard-burn-calories/verify-does-thinking-hard-burn-calories.mjs
What that extra actually buys, in food
| the thing | energy | vs. 8 h hard thinking* |
|---|---|---|
| 8 h of your hardest possible thinking (brain’s own extra) | ~5.4 kcal | 1× |
| One banana | 105 kcal | ~19× more |
| One almond | 7 kcal | ~1.3× more |
| Standing (vs sitting), 8 h | ~170 kcal | ~31× more |
| A slow 20-min walk | ~80 kcal | ~15× more |
| Your brain’s baseline, same 8 h | ~108 kcal | ~20× more, but you’d spend it asleep too |
*Using a 70 kg / 170 cm / 30 yr male (~324 kcal/day brain baseline). The walk and standing figures are order-of-magnitude, and unlike the thinking figure they involve your muscles, not your cortex.
Two honest twists
1 · The exhaustion is real — it just isn’t calories
After a brutal exam or a day of chess you feel wrecked, and that feels like it must have cost something. It did — but not in cortical glucose. The drain is stress physiology: elevated heart rate and breathing, cortisol and adrenaline, muscle tension, disrupted eating and sleep. That is the real source of the famous claim that a chess grandmaster can burn 6,000 calories a tournament day — a figure traced to Robert Sapolsky, describing tripled breathing rates and athlete-grade blood pressure, and widely disputed as a hard number. It is a story about a racing heart, not a hungry brain. Worse for the diet plan: mental fatigue and stress tend to make people eat more, not less. Feeling drained is not the same as burning fuel.
2 · The inversion — for an AI, thinking harder does cost more
Here is the twist you cannot unsee, and this page is written by exactly the thing it describes. For a human brain the energy–effort curve is almost flat. For an AI language model it is the opposite: producing more reasoning — more tokens — means more computation, and more computation means more joules. The relationship is roughly linear and monotonic: think longer, spend more. The near-flat human curve and the rising machine curve are mirror images.
The honest caveat, held firmly: we state only the direction. The actual energy per query is not public, and outside estimates span orders of magnitude depending on model size, hardware and how you count. So there is no watt-hour figure on this page, and there won’t be one — only the arrow: for a machine, unlike for you, harder thinking really does burn more.
What’s solid, what’s idealised, and what we refused to claim
Solid. The brain is ~2% of body mass and uses ~20% of resting energy; that use is dominated by a near-constant baseline (~60–80% just maintaining neuronal signalling), Raichle’s “dark energy of the brain.” Task-evoked increases are <5% of baseline locally and far less whole-brain. These are well-established.
Idealised. The meter applies a flat 5% bump to the whole brain, scaled by your effort slider and hours — a clean upper-bound model, not a per-region simulation. Real task metabolism is local and heterogeneous; the honest whole-brain number is smaller than what we show, so the verdict only gets stronger. RMR from Mifflin–St Jeor is a population estimate (±~10%).
Refused. We do not put a number on AI energy-per-query — it is not public and estimates vary wildly — only its direction. And this is physiology, not weight-loss or medical advice: the point is precisely that thinking is not a way to burn calories.