Hmm, I realize that some monks might object to that, but do we really need to use Text::CSV when the whole thing can be done in 4 lines of actual code?

use strict;
use warnings;
use feature qw/say/;

my %results;

while (<DATA>) {
    my ($id, $string) = (split /[,\s]+/)[0,1];
    next if defined $results{$id} and length $string > length $results
+{$id};
    $results{$id} = $string;
}
say "$_ $results{$_}" for sort keys %results;

__DATA__
A, texttexttext, col3, col4,
B, textt,        col3, col4,
A, text,         col3, col4,
B, texttex,      col3, col4,
[download]

Or, possibly one more code line if we really want to cache the length in the hash:

# ...
while (<DATA>) {
    my ($id, $string) = (split /[,\s]+/)[0,1];
    my $cur_len = length $string;
    next if defined $results{$id} and $cur_len > $results{$id}{len};
    $results{$id} = { str => $string, len => $cur_len };
}
say "$_ $results{$_}{str}" for sort keys %results;
#...
[download]