Why is it a middling product? Most clones don’t approach its performance, and it solves a specific problem well in a way that was awkward and ignored by most frontier labs.
If I seriously needed a classifier, I would just train my own and it would run on an iPhone. Any CTO worth their salt would suggest the same, because it's not even remotely comparable to training a large language model (w.r.t. compute or training data required).