Supporting at least the subset of types of the form Tuple[T1, T2, <some Ts>, TRest, ...] — that is, with ... allowed only in final position and only once (specifically not including things like Tuple[int, ..., str, bool] or Tuple[str, ..., int, ...] which too can have their merits) — seems more or less reasonable to me. (Seen this kinds of types in Ceylon language, though there tuple/tuple-like types are intristically linked with callable types, and there is more reason for these that may be here.) Not including the kinds of types in parentheses should ease typechecking, as far I understand it.

I see at least one weak use in DSL design: writing values of the allowed type Tuple[str, List[int]] (contrived) is a bit more tedious than writing values of Tuple[str, int, ...] instead. Maybe there is even a non-contrived example. :)

It can be argued that if tuple concatenation types for the tuple types postulated today, will become typable in e. g. MyPy or other major typecheckers, then the proposed kind of tuples may be a pain to cover.

Thanks @arseniiv . Just to add to this, my use case is a function in a class which returns a Tuple[Foo, Any, ...].
This class is meant to be subclassed and the function overridden. All subclasses have to return Foo as the first element in the Tuple, but can choose to add 0 or more additional elements, depending on the subclass usage.