Every steak recipe says “three minutes a side”. Heat travels from both faces to the middle, and that takes time proportional to the square of the thickness — so the one number every recipe leaves out is the only one that decides the answer.
You cannot read it off the dial — the dial sets power, and the pan sets temperature. Two ways to find out:
Landmarks. Butter browns around 150 °C. Oil shimmers and thins at 180–200. A flicked drop of water breaks into beads that skate across the surface (the Leidenfrost point) from about 200. Most seed oils smoke around 230, and cast iron left dry will haze at 250–260. If it is smoking hard before the steak goes in, you are past searing and into burning.
Rough dial settings, for a heavy pan preheated 4–5 minutes. These are starting points, not measurements — hobs vary enormously:
| Hob | ~200 °C | ~230 °C | ~260 °C |
|---|---|---|---|
| Gas | medium | medium-high | high |
| Induction (1–9) | 6 | 7 | 8–9 |
| Electric radiant (1–6) | 4–5 | 5–6 | 6, and slow to change |
An infrared thermometer costs little and removes the guesswork entirely. Induction responds in seconds; radiant electric takes a minute or more to catch up, which is why it is the easiest hob to overshoot on.
Rest time is advice, not a simulation result. Resting is mostly about letting the fibres relax and reabsorb juice, which is not a temperature you can read off; the figure above is the usual guidance for this thickness. The simulated rest is stopped at the moment the middle reaches your target, because past that point it drifts towards the steak's mass-average, which is energetically right and not a thing anyone waits around to watch.
The rest is simulated, and the pull point is solved against it: the page searches for the moment to take the steak out such that after the FULL rest the middle lands on your target. That is why the pull temperature is well below the doneness you asked for. Scrub past the pull and you can watch the gradient collapse inwards.
Runs the same simulation forward in real time. Sound needs one tap first.
Low oven first, pan last. The page used to say it could not answer this. It can: an oven phase is every face exposed to hot air, and the sear is the ordinary pan cook starting from the field the oven produced. The claim worth testing is not that it is quicker — it is emphatically not — but that it is more evenly cooked. Since the solver already records the peak temperature every part of the steak ever reached, the overcooked band can simply be counted.
A conventional oven — still air plus radiation from the walls, which works out near 18 W/m²K and is evaluated once rather than tracked as the surface warms. Roughly half of that is the walls glowing, not the air. A fan oven moves considerably more heat and will beat these times; a crowded oven will not.
Heat enters from both faces and has to reach the middle. That is conduction, and conduction time goes as the square of the distance — so thickness is the input that decides the answer. Weight tells you how many people it feeds and how far the pan sags when it lands; it moves the time by under a tenth. Every recipe that says “for a 250 g steak” is anchored to the wrong number.
When a steak leaves the pan the outside is far hotter than the middle, and that heat keeps moving inwards. If you take it out at the temperature you want to eat, you will eat it one doneness further on than you planned.
How far it drifts depends on how hard you seared it, and on how thick it is — and at one end it does not drift up at all. The usual published figure is 3–6 °C, measured on a gentle cook. This page simulates the rest rather than assuming a figure, and on cast iron at 230 °C it gets anything from −2 °C to +23 °C depending on the steak.
An 8 mm steak actually cools while it rests: there is not enough heat stored in its outside to beat what it loses to the room, so you pull it above the temperature you want. A 40 mm one gains more than twenty degrees, because a hard sear leaves roughly a third of the meat past well done and that shell has to put its heat somewhere. Both figures are shown, labelled, under How sure is this answer? — sear gently and the published one is the right one.
The single dramatic flip is a photography technique. Turning every 30 seconds gets to the same core temperature sooner and with a narrower band of overcooked meat under the crust, because neither face spends long losing heat to the room. You trade some of the grill-mark contrast for a more even inside.
A thin pan loses more surface temperature to a cold steak than it can replace, and the meat ends up steaming in its own moisture instead of searing. That is why cast iron is worth the weight, and why a non-stick pan — whose coating should not go past about 230 °C — is the wrong tool for a steak.
Everything here is a model of a steak, not your steak. It does not know how flat yours is, how wet the surface is, or what your hob actually does. Use these numbers to know when to start checking, and let an instant-read thermometer make the decision.
Conduction, evaporation and the crust are modelled. The surface carries a finite amount of water; while it lasts, anything arriving above 100 °C is spent boiling it off rather than heating the meat, and as the face dries its contact with the pan falls away — a crust is porous and warped, and stops conducting. That is why the surface settles in the 140–160 °C range where beef actually browns.
Water keeps leaving after the free water has gone, wicked out of the meat, which is what you are watching when a rested steak steams. That is modelled too, at the lesser of what the air can carry away and what can wick through a crust — set so the steak loses the 1–2% of its mass a resting steak really does lose. It is the reason the rest is not simply the heat being shared out.
What is not modelled: fat rendering, a bone as a separate material (treat T-bone times as the meat-only answer), a steak thicker at one end than the other, and how flat yours sits. Times are a starting point for the first steak; an instant-read thermometer is the only thing that knows when yours is done.
The page solves heat conduction across the steak in two dimensions — through the thickness and across the width — as an explicit finite-difference solve of the heat equation on a grid of 2,173 cells, stepped forward a fraction of a second at a time. It is not a table of recipe times with your inputs looked up in it.
Each face is either against the pan or exposed to the room and they swap on your flip schedule, while both side edges lose heat to the room the whole time. Solving only through the thickness, as this page first did, makes every point at a given depth identical by construction — so “uneven” could not exist, and the edges that actually overcook were invisible. The pan side uses a contact conductance of roughly 1100 W/m²K for cast iron; the air side uses about 20, which covers free convection and radiation together. The steak’s own conductivity, density and specific heat come from the cut, because fat conducts heat more slowly than lean muscle.
The deliberate choice is not to use the usual one-term (Heisler) approximation. That is only valid once the Fourier number passes about 0.2, and a searing steak spends most of its time below that — precisely where the shortcut would be wrong. Solving it directly costs a few milliseconds and has no validity window.
A 25 mm sirloin from the fridge, cast iron set to 230 °C, turned every minute, cooked to medium rare. The surface under the steak settles near 211 °C once the cold meat has pulled it down. The centre reaches 43.9 °C after about 7 minutes 15 seconds in the pan — and that is the moment to take it out, because over a 5:50 rest it drifts up 12 °C to land on 56.
That pull temperature looks far too low, and it is the part worth dwelling on. It is not an offset taken off the target: the page simulates the whole rest from every candidate moment and picks the one that finishes at 56 °C. On this cook the steak already averages 63 °C when the centre is still at 44 — the outside is carrying the heat, and the rest is what shares it out, minus what leaves as steam.
Cut the same steak 40 mm thick and it needs about 13 minutes 15 — not the 60% longer its thickness suggests, but well over twice as long.
Because it is still cooking. The outside of a seared steak is far hotter than the middle, and that heat keeps travelling inwards after it leaves the pan.
The page does not subtract a fixed carryover figure to decide when. It simulates the whole rest from every moment the steak could come out, and picks the one that finishes on your target when you cut it. That is why the pull temperature can look far below what you asked for: on a hard sear the outside is holding most of the heat, and the rest is what moves it to the middle.
Hardly at all for timing. At a fixed thickness, going from 6 oz to 14 oz changes the time by about 8% — it is the same distance from the face to the middle, and the heat does not care how wide the steak is. Weight does affect how far the pan temperature sags when the steak lands, which is modelled, and it decides how many people you are feeding.
A face that is not on the pan is losing heat to the room. Turning often keeps both faces near the pan temperature on average, so more heat goes in per minute and the gradient through the steak is gentler. The cost is that you do not build the same dark crust in one unbroken contact, which is what the single flip buys.
Pick the closest by thermal mass. A thick disc-base stainless pan behaves more like carbon steel than like a thin one; an enamelled cast-iron pan behaves like cast iron. The setting that matters most is the surface temperature, and the only reliable way to know yours is an infrared thermometer.
Yes — there is a section for it above. An oven phase is the same conduction problem with every face exposed to hot air instead of one face on a pan, so the sear that follows simply starts from the field the oven produced, dried surface and all.
The result is more interesting than “it is better”. The two methods cross over at roughly 30 mm. Above that the reverse sear is clearly more even: at 45 mm about 24% of the steak overshoots medium well, against 48% from the pan. Below it the reverse sear is worse — at 22 mm it is 51% against 40% — because a thin steak has no deep gradient to protect and the long oven dwell just holds more of it hot for longer. For a thin steak, turning it often in a hot pan is the more even cook, and it is done in a fifth of the time.
Two-dimensional conduction through a rectangular cross-section — heat enters from the pan face, leaves the upward face, and leaves both side edges, which is what makes the rim and corners run ahead of the middle. Uniform meat, with a pan surface that sags on contact and then holds. Evaporation from a wet surface and the loss of contact as a crust forms are both modelled. Not modelled: fat rendering, bones, or a steak that is thicker at one end than the other. Times are a starting point for the first steak, and an instant-read thermometer is the only thing that knows when yours is done. Cooking times are guidance, not food-safety advice — if you are cooking for someone pregnant, elderly, very young or immunocompromised, follow your local food-safety guidance on internal temperatures instead.
Releases in which this page changed, newest last. Derived from the archived copy of every release, not from notes written afterwards — so it reflects what actually shipped. Site-wide passes are left out; they are in the full changelog.