Biological information has several hierarchies is ambiguous between syntactic information in a hierarchical modular system, and functional information, without a particular context.
Out of necessity the DNA is sequence is generally highly conserved, but small changes can easily lead to gross changes in folding, function and the minimum energy states of the protein. (Proteins are ultimately dynamic systems, with many constantly sampling their minimum energy conformation structures - the static pictures on magazine and book covers are misleading)
Additionally an allele, by definition, is a gene variant that occurs in at least 1% of the population - by consensus (and and sufficient statistical power). As such the term is part of population genetics. For instance, for a 400AA long protein, that would mean that a particular homologous recombination event has to occur very frequently and be favored by at least one of many (known) mechanisms.
Is new genetic information ever created during sexual reproduction, or only through mutations during an organisms life time?
Yes. Recombination has several mechanisms, is not a perfect event, and shows different efficiencies depending on the type of mechanism and "helpers"/helper-types as well as helper concentrations involved, - notwithstanding cofactor-concentrations.
John Collier, Hierarchical Dynamical Information Systems With a Focus on
Biology, Entropy 2003, 5, pg. 100-124