Assuming that there will be more than one line of data to process, I wrote the following:

use strict;

{
  local $/ = 'output';

  while (<DATA>) {
    chomp; # drop 'output'
    my (undef, $line) = split /input/;
    $line =~s/,/;/g;
    print "$line\n";
  }
}
__DATA__
input CK, n3065gat, n3066gat, n3067gat, n3068gat, n3069gat, n3070gat,n
+3100gat, test_si, test_se;
  output n3104gat, n3105gat, n3106gat, n3107gat, n3108gat, n3109gat, n
+3110gat;
input CK, n3065gat_2, n3066gat_2, n3067gat_2, n3068gat_2, n3069gat_2, 
+n3070gat_2,n3100gat_2, test_si, test_se;
  output n3104gat_2, n3105gat_2, n3106gat_2, n3107gat_2, n3108gat_2, n
+3109gat_2, n3110gat_2;
input CK, n3065gat_3, n3066gat_3, n3067gat_3, n3068gat_3, n3069gat_3, 
+n3070gat_3,n3100gat_3, test_si, test_se;
  output n3104gat_3, n3105gat_3, n3106gat_3, n3107gat_3, n3108gat_3, n
+3109gat_3, n3110gat_3;
input CK, n3065gat_4, n3066gat_4, n3067gat_4, n3068gat_4, n3069gat_4, 
+n3070gat_4,n3100gat_4, test_si, test_se;
  output n3104gat_4, n3105gat_4, n3106gat_4, n3107gat_4, n3108gat_4, n
+3109gat_4, n3110gat_4;
[download]

It works by setting the input record separator '$/' to 'output', so every line you read will be terminated by 'output', rather than by the usual EOL.

chomp removes the 'output'.

Then I split the line just read on 'input' and throw away everything before 'input', leaving me with nothing but what is between 'input' and 'output'. The substitution s/,/;/g takes care of changing commas into semi-colons. There is no need to quotemeta them as they are not special in a regex. The result then gets printed.

BTW: you are opening, printing and closing the file for every line. Not only is this very slow, it is also wrong as each opening of the file for output with '>' will delete everything which was in that file, so you end up with only the result of your last print. Keep your open and close outside of the loop.