It's interesting to compare it to an encoding like base65536 [1].
From that respect, alphabinary feels less like an encoding and more like steganography, especially given how low efficiency the encoding is overall. Given that steganography of a sort seems at least partially an intended goal, that's not necessarily a criticism here, just an interesting observation.
If my goal were just to make sure I can encode binary things in a tweet, the current best bet appears to be base65536 (with the dream of Unicode filling out enough to support a good base131072), but alphabinary seems like an interesting idea to keep in mind for possible steganographic uses where one might want to make it less obvious any encoding was used at all.