Radiator health in Service Mode: where the percentage comes from and why it falls

What this line is

Cars on firmware from the 2026.20.* branch have a radiator health percentage in Service Mode, on the Thermal tab. What lived in that spot before was two values, 100 or 0, and there was not much use in them (#476758). Now the number is fractional. In a single week of July 2026 the chat reported 79% on a refreshed Model 3 (#476758), 83% (#477353) and 76%, which became 73% a month later (#477359, #478071).

Tesla has not published a formula. From the way it behaves the reading looks like a computed estimate: the car compares actual heat rejection with what it expects at the current fan speed and loop temperatures, and shows the ratio as a percentage. This is the chat’s version, it has no official description behind it, and it deserves exactly that much trust.

The line lives in Service Mode. How to get in there and what else is useful inside is described in the write-up on Battery Test.

What happened in the summer of 2026

The numbers went down for a lot of owners at once. One watched a drop from 100% to 65% over about three weeks, with no crashes, faults or work done to the car (#469940, #469942). Another car went from 98% to 83%, which was read as a stack packed with debris (#449903, #449905).

On non-original stacks the picture is sharper. 89% turned into 0% in a week or two (#484871). An owner with a Chinese radiator saw 83% become 78% over two months (#484884). A build of an aftermarket stack plus the original fan held at 84-86% (#484887). The obvious explanation is that the aftermarket part sheds heat worse, the car sees the gap against its calculation and writes it down in percent.

The odd part: the reading does not come back

This breaks the whole “read the number, wash the stack, check again” routine. After the stack is cleaned the percentage stays where it was. After a new radiator goes in, the same thing: the owner of a pre-refresh car lived with the new part for a month and a half and saw no change, while the self-test and the thermal test both passed (#469952, #469984). The red indicator dot goes out with a delay after a wash, and sometimes it does not go out at all (#449926, #449933).

From this comes the practical part. One percentage seen once proves nothing. What works is the trend on one and the same car, and even that comes with caveats.

When it really is about dirt

Rely on the signs you can check with your hands and eyes.

Fan speed on a warm car. A steady 3000+ rpm with no obvious load means the stack is getting clogged (#484889). Summer consumption: after a proper cleaning it falls by around 30%, and that shows up on the consumption graph (#464965, #464968).

Faults. VCFRONT_a611_poorRadiatorHeatRejection turns up in hot weather, when refrigerant temperature goes past 80 °C and the compressor spins up to 9000 rpm (#455493, #457102). A separate signal comes from the fan: the system asks for 900 rpm, gets 100, and the noise level is that of a vacuum cleaner (#465008, #465536).

Behaviour on fast charging. A live case: a Model Y Performance 2023 dropped its DC session three times in a row at 40-50% charge, Service Mode showed the FSD computer reaching 96 °C, and a fault for low fan speed was hanging there as well (#477907, #477942, #478009). The chat’s advice was blunt: no fast charging until the radiator has been inspected, or you can cook the computer (#478010).

What to do with your own car

Look at the number and note the date. Look again a month later. After that watch the fan speed and the consumption: if both are normal, put the percentage aside and get on with life.

If there are signs of overheating, the answer is a cleaning, and it has to be done properly. Vacuuming through the grille achieves nothing, the bumper has to come off: on Model 3 and Model S there is no other way in, and one owner spent at least an hour on the whole job (#449926). The order of work, the tools and what you must not do to the fins are in the article on cleaning the condenser and radiators.

If it comes to replacing the stack, keep the spread in price and quality in mind. A Chinese stack for a Model Y 2022 cost around 250$, a used original around 500$, and owners running aftermarket parts regularly catch cooling and louvre faults in hot weather (#471847, #471864). Take a stack for your own model only. For a shop that does this with the bumper off, look in the directory.

What we do not know

The calculation formula. The averaging period. The conditions under which the value is recalculated upwards. The list of firmware where the reading behaves the same way. While all of that stays closed, treat the percentage as a hint that you verify by other means.