[quux]

It must, if it now outputs more tokens than 4.6's tokenizer for the same input. I think the announcement and model cards provide a little more detail as to what exactly is different