I haven't checked how fast this is but you could try using Hail (disclaimer: I'm a maintainer). You'll need to install the GCS connector first to read from Google Cloud Storage:
curl https://broad.io/install-gcs-connector | python3
import hail as hl
mt = hl.import_vcf('your.vcf')
mt = hl.split_multi_hts(mt) # use the same biallelic representation that gnomAD uses
mt.write('data.mt') # store in Hail's native format for faster queries
mt = hl.read_matrix_table('data.mt')
sites = hl.read_table(
'gs://gcp-public-data--gnomad/release/3.1.2/ht/genomes/gnomad.genomes.v3.1.2.sites.ht'
)
mt = mt.anti_join_rows(sites)
That'll filter down to just those variants (defined by chromosome, position, and alleles) which are not present in gnomAD v3.