I think that the question is difficult to answer in the current phrasing. However, I will show how it might be easier to answer by filling in some blanks and defining some terms.
I am going to define as an "allele" any nucleotide difference between two sequences. This is the molecular genetic definition more or less. I am further going to interpret "mismatch" as also including sequence gaps, as in the case of indels your alignment will include gaps. I am assuming that you are talking about aligned sequence differences, because you mention alignments. I am going to ignore the focus on "genes" because at the molecular level visible in alignments, alleles are not necessarily tied to annotated genes (even though they are usually introduced in that context).
So in this case, if you assume that most alleles consist of single nucleotide variants (probably not true, but a helpful simplifying assumption), you can deduce that nearly all "alleles" thus defined will differ from each other at a single nucleotide position. A larger and more complex genotypic difference (for example a microsatellite variant or a transposable element (TE)) will involve a proportionally larger number of mismatches (i.e. gaps).
Indeed, if we look at the wikipedia page for "allele":
At the lowest possible size an allele can be a single nucleotide polymorphism (SNP).[7] At the higher end, it can be up to several thousand base-pairs long.[8][9]
You might however mean something different. It's possible that the question is rather something like this:
How much segregating genetic variation of an organism (say humans) consists of single nucleotide variants (SNVs) vs. larger/more complex variants?
That's a somewhat different but much more answerable question. It is unfortunately limited a bit by our ascertainment bias in favor of SNVs, which are just easier to find. But the best estimates that I'm aware of suggests that there are probably slightly more microsatellite variants than SNVs, and some work suggests possibly a similarly large number of large-scale repeat variants (such as TEs).
I can imagine yet another interpretation, which might be something like:
If we align two (or 10, or however many) homologous sequences of defined length drawn from a natural population of an organism (say humans), at how many positions can we expect the aligned portion of any pair of sequences to differ per base pair?
This is more straightforward: the alignable portion of the human genome is expected to show approximately 0.1% differences. However, this is pretty much only counting SNVs. This will vary from species to species. Off the top of my head, I remember that interfertile Arabidopsis thaliana strains can differ by anything from 0.1% to 1% per aligned base.
