With read(), the return value is separate from the data read. It is arguably elegant to read as many characters as possible and to not consider hitting the end of the file as an error—just a point that can't be read beyond, like a stop on a piece of machinery.

With getchar(), the decision was made to return the char directly (to have used a pointer to a single char would be daft, from the point of view of economy) and to use an in-band code to indicate end of file. This "in-band signalling" is considered a bit of a kludge by some people. So I'd say read() is a better/cleaner design.

See https://en.wikipedia.org/wiki/Semipredicate_problem for some related thoughts.