Is the HP 50g the last RPN calculator?

[3298]

Purely postfix? Nope.
There are commands that take parameters from the runstream. In SysRPL they are everywhere, in UserRPL they are relatively rare but still present. The most obvious UserRPL examples are FOR (takes a local variable name after it which is used as loop counter storage) and -> (takes as many local variable names as it can get, and creates local variables with these names and values from the stack). There are some other less obvious examples like the command inserted by the UserRPL compiler when it encounters a quoted name (as opposed to an unquoted one, which would try to recall and execute the variable's contents): this command takes the object after it (i.e. the variable name) and puts it on the stack. That's how UserRPL arranges for quoted names to get put on the stack without getting evaluated.
In a broader sense, many of the control flow commands in UserRPL could be described as taking parameters from the runstream: for example, THEN takes the two objects after it and discards them (that is, it skips them), if the number it has taken from the stack is not equal to 0 (the first object is the code block up to the corresponding ELSE, or if that is not present, up to the corresponding END, wrapped by the compiler into a program using invisible delimiters; the second one is the ELSE or END itself).

So that argument doesn't work. Besides, using "purely postfix" as a criterion for not being RPN (Reverse Polish Notation, a.k.a postfix notation) is a bit awkward, don't you think?
It is a bit unfortunate that the programming language of HP's keystroke-programmable RPN calculators is often also called RPN (does it even have an official name?), but I guess it's too late to fix that.