For search , one of the following options must be specified:

-fast
-sensitive
-verysensitive

The names are pretty self-explanatory. For most applications, I would recommend ‑sensitive unless the ~3x greater cost in CPU time is important.

To give you an idea of their relative execution times, on the SCOP40 test , elapsed times on a 32-core i9-14900K Linux server were

fast = 49 secs.
sensitive = 3 mins.
verysensitive = 14 mins.

With verysensitive , the output file size is typically much larger because the number of reported hits is much larger (assuming you use the default E-value threshold ). On the SCOP40 test output file sizes with -columns query+target+evalue were:

fast = 10 Mb.
sensitive = 15 Mb.
verysensitive = 2.9 Gb.

For most applications, the large majority of ‑verysensitive hits will be false positives; this option is included for studies where it is important to find alignments having only very weak structure similarities which probably do not indicate evolutionary relationships or similar functions.