Since sentience boils down to having at least one sense (input) and being able to make sense of it (make a model of it), it seems to be edging into yes.
An analogy I make with things in this field is that a human equivalent to these is someone with a super talent in remixing anything digital, but that is developmentally stuck as a three year old.
From what I recall of earlier work in the field, the three year old equivalency seems to be where things have been for well over a decade. And still is stuck there.
Making LLMs might be a research track to see if that gives people ideas in how to get past the three year old equivalency, but I fear the industry has widespread belief that LLMs will magically directly push past the stall.