No, it's pretty easy to get better sensors than that. E.g. a cheap-ass SHT20 will do +-0.3C. In fact, my own automotive parts recommendations are the next grades up (SHT21/SHT3x) as standard for my employer's boards because the cost difference is justified.

Never underestimate the ability of a manufacturer to select subpar parts to save 25 cents on the BOM and spend 6 figures elsewhere trying to fix the resulting issues though.

No, you read the datasheet wrong. It's +/- 3C.

I'm looking at the datasheet right now. Relative humidity accuracy tolerance is +-3.0, temperature is +-0.3. Different tables.

I imagine the temperature gradient across the car might be +/- 3degC, ignoring the actual sensor.

But the weird part is that OEM seem to be fine with that variation when it comes to climate control. Do they use multiple temp sensors?

Measuring temperature is not trivial. There's convective, conductive and radiative heat transfer and they all factor into the measurement. And that may not accurately reflect the "cabin temperature," particularly in a parked car.

It's undoubtedly more than that, depending on where you measure. The gradient between e.g. the roof lining and the AC vent could easily be 20C+ degrees on a hot day. Most boards on a vehicle will have their own temperature sensors to measure enclosure temps, and there will be zonal sensors at various points in the cabin as well. The climate control loop will be defined in terms of those zonal sensors.

I think the problem is not the accuracy of the sensors themselves but that its difficult to place them inside the car without influencing the measurement.

Depends on the sensor, depends on its calibration (or its lack of such functionality). Often a function of cost and/or size, as well as the means by which it measures temp.

Scientific sensors are highly accurate and can also be small, but you have a steep cost increase of course.

If you just buy a bare sensor, yes, it's going to be +/-5. They also have a non-linear response which needs to be dealt with as well.

If you are only concerned about a 20 or so degree temperature range it's not an issue, but if you are trying to read over a 100 degree range, you'll want to account for non-linearity as well.

Also, accurate and precise to 10ths of a degree isn't really attainable unless you do fancy math as the sensor will heat each time you read it. The idea is to take multiple readings and average them but unless you are accounting for the heating of the sensor, your numbers will be garbage.

This is for consumer grade sub $50 sensors. Of course you can go fancier but you have to pay for it.

This isn't something I know anything about, but I know that 1-Wire exists and so on so am able to locate something like https://www.analog.com/en/products/max30207.html pretty easily. $2 in quantity, reports a temperature digitally, accurate to 0.3 C between 0 and 70 C.

What is it about that device (or similar) that would put it out of scope?

It's probably not that the sensor is bad. It's the location of the sensor that is tucked away in the interior foam so that it's not reading the air inside the cabin