The attribute "binary" may sound a bit confusing when referring to content as all content stored on a computer is binary. We use the term here for arbitrary content which can be stored on the platform and retrieved as binary identical copy. For example if an RDF/XML file is uploaded as binary content one will be able to retrieve a copy that is identical to the original file, i.e. including comments and possible syntax error, would the same file be uploaded as graph (rather than binary content) the file would be parsed on upload (producing an exception in the case of syntax error) and one will be able to access a serialization of that graph in any supported RDF format, the downloaded documents will describe the same graph but not (necessarily) be binary identical to the uploaded document.
