I contacted Affymetrix directly, and they recommend to always use the latest version of the annotation file for the appropriate array.
So for my case, that is the annotation file for GenomeWideSNP_6 Annotations, CSV format, Release 35 (313 MB, 4/30/15) (current latest release). The Annotation file contains the SNP positions. So Liftover is not required here.
Please know that the reason why you see different versions of the
annotation file is because we design our arrays using information from
public databases, however these databases are constantly updated and
annotation updates incorporate current releases from GenBank, RefSeq,
Ensembl, UniGene, Entrez Gene, UniProt and UCSC, as well as sequences
from other organism-specific databases;
Some of our arrays were designed
several years ago at which time the information on the databases was
different and most of our arrays were last updated in April 2015 so
some previously entered genes have not been validated over the years
therefore taken out of the database but our probes on the arrays are
still designed to target those.
As more genes are discovered, and others are removed if they fail
validation by the scientific community, the number of gene assignments
in the annotation files will fluctuate. The "well-annotated" gene
counts are for gene symbols that are associated with transcripts
having the best experimental support, such as the reviewed and
validated records of RefSeq or those with accessions beginning with
So you should always consider the Current NetAffx Annotation File
only, to analyse your data, regardless of how old or when your array
was processed, since it will give you the most up to date annotation
Please know that regardless of the annotation variation, all the
probes are always in the same array coordinates, this is determined by
the library file, which is unique for each array;
If we add or remove probes we will change the array name, has it
happended several times, but for each array, regardless of the
annotation version, all the probes are in the same place;
This is valid for our hole array portfolio, so SNP6 also.