There is no "likelihood". The mismatch value is used to compute a score. Read up on alignments in general.

The magnitude of the mismatch should be interpred in context of all other scores (for match/and gaps), the value alone it is not meaningful. For example, if the alignment hits a mismatch but if after the mismatch it can be extended to produce two more matches then it will continue over the mismatch. If not it will stop at the mismatch and will clip your alignment right there.