After testing, it seems to me like Qwen Image Edit is to Kontext like what Qwen Image is to other T2I models.
It's a huge amount better for more complex prompt following, but it has slight tunnel vision on what it wants to do and in what way, and the main problem: it's only good for destructive edits.
For a lot of use cases where you want the image to stay mostly the same aside from the slight VAE quality loss, it looks like it's quite noticably worse.
For example, no matter what, I can't reproduce anything similar like in picrel that was possible with Kontext, none of the below prompts work, and they also make her hair brown and change her eyebrows completely by default
"elongate her hair"
"make her hair longer"
"make her hair long without changing anything else in the image"
"change her hair to be very long while keeping the rest of the image as is, keep her eyebrows the same"
It's fun to play around with given the much, much better complex prompt following capabilities but it's not good enough for iterative work over existing images that should keep most of their look. Granted, Kontext isn't that great either given the censoring and limited things it can change, but when it works, it works very well.