I take it you have a BAM file ?

Freebayes is robust to many technologies but needs a lot of filtering on quality etc. I would recommend that and bcftools for filtering - but I have not worked on Ion torrent data, which is somewhat rare in the community.

GATK is customized for human data. You did not note what genome you are working with, so it makes the question a bit more complicated.

In my testing, Freebayes performs abysmally, so I can't second that recommendation (but at least it is easy to use, unlike GATK). I am biased, since I wrote it, but I recommend BBTools' callvariants.sh for variant-calling. It's calibrated for Illumina data and I have gotten a report that there were some false-positives in Ion Torrent data, so I can assure you that it is not perfect. However, I would be willing to work with you if you want to help me calibrate an Ion Torrent mode. JGI doesn't have any Ion Torrents so I have not calibrated any software for them (we use Illumina and PacBio). You can use callvariants.sh as it stands if you want and it will still perform well on everything other than homopolymer indels, where it may give some false positives for Ion Torrent (or Nanopore) reads.