I guess that strongly depends on the use case.

In programming, they are far from good enough.

In article writing too. Now we can distinguish quickly a text written by a human from one written by a large language model.