The filter I use, because the real problem is volume rather than any single one: where in the process does it arrive?
Before any human conversation, decline or ask to move it later. A take-home as a screening filter is asking for a weekend from strangers, and it is the version most likely to be a black hole. It also tells you the process is designed around their convenience.
After a first conversation, before the final stage: this is the normal and defensible position and I do these.
As the final stage, with the team, discussed live afterwards: the best version. The discussion is where the value is for both sides, and it means the code will actually be read by someone who talks to you about it.
So the questions worth asking before starting: how many candidates are at this stage, will I get feedback either way, and will we discuss it with the team afterwards. Reasonable companies answer all three.
And be willing to spend the budget unevenly. Two take-homes at real effort for jobs you want beats five at half effort. A submission that is obviously going through the motions is worse than not applying, because it costs you the weekend anyway.