JLAtg always with long /ā/, but in the tradition of the Genizah PTA texts with short /a/ when standing alone, (i.e. in a closed short syllable in construct state), /ā/ with pronominal suffixes. This latter tradition is followed by some LJLA pointed manuscripts. See the similar pattern with yt prep.

This preposition otherwise occurs with non-personal objects only relatively late, thus making the reconstructed form from OfA Ahiqar dubious.
With verbs of speech in translation and later Syriac, almost certainly calqued from the Greek, as argued most recently by Butts, Change p. 144ff.