In my opinion, the most accurate ones are Originality (paid, only a few credits for free) and Gowinston.ai (the same limitation with Originality applies), along with GPTZero and Copyleaks.
I scanned one of my posts which has been flagged by Sapling.ai as 100% fake (AI generated) in Originality (free version) after you praised this detector, to my surprise it is also showing my post as 94% AI (almost same as sapling.ai)
While with GOwinston.ai (free version) it is showing right results : 0% AI (I know it is right because I wrote the post lol), along with Copyleaks (0%) and GPTzero (0%).
In my opinion, the combo of Copyleaks+GPTzero is very accurate.
I have scanned a few more posts, and Originality is giving me results similar to those of Sapling.ai.
I think some detectors are just using perplexity in order to detect AI but the right method is to compare it with previous human written texts + perplexity, which is exactly what Copyleaks and GPTzero are doing.