Hello!
Negative HOMO-LUMO gaps, which are OK within the DFT model used, do occur from time to time. In this case, the gap is quite large, however, so I suspect it has converged to an excited state. (I haven't played around with the empirical dispersion correction too much, so perhaps larger negative gaps are more common with those, for some reason.) There's (at least) two things you can do to try to fix that:
1) Tighten the $scfconv and lower $orbitalshift values and restart. Because of those, you might need to increase the $scfdamp value, too. This could help.
2) Manually edit the "mos" file, and move orbital 105 data before orbital 104, change labels, and restart. This usually helps.
Have a nice day,
Mikael J.
http://www.iki.fi/~mpjohans