As varro says, the forms are reconstructed with a long vowel, as mēd, tēd, sēd. AFAIK there's no direct evidence for the vowel quantity. The reason for the long-vowel reconstruction is that there seems to have been a sound change between Old Latin and Classical Latin in which final -d was lost after a long vowel; this is the reason for the final long vowel of ablatives in -ā (< -ād), -ō, (< -ōd), etc.
The loss of -d is thought to have occurred only after long vowels. I'm not sure how certain we can be about that, though. Old Latin did have -d after short vowels, resulting from word-final voicing, e.g. FECED = fecit, and it's true that these were not lost. But it's also true that these -d were replaced by -t (perhaps analogically) at some point, and if that happened before the -d-loss sound change, then these forms aren't evidence either way.
As Sihler mentions, Sanskrit has short vowels in the corresponding forms (mat) etc., but as he says, it's plausible that these could have been lengthened in Latin by analogy to the noun declensions, so that also isn't too helpful.
There may well be other relevant evidence I'm missing, but it seems that the answer is "probably the vowel was long but we can't be sure".