That said, operations that need to see everything (sort, group by, stats) do buffer in memory. For those on massive files, `first N` or `where` before the sort helps.

Honest answer: I haven't benchmarked against jq (and wouldn't call it a replacement) on multi-GB files yet. jq is C, jsont is Go. For pure filtering speed jq probably wins. The value is in the explore commands (schema, find, stats) that jq simply doesn't have. and the way agents can interact with this cli which should save you/them 10 minutes of guessing on a file you've never seen.