Published catch rates for AI code review tools range from 6% to 82% on one vendor-run test of 50 bugs, and each vendor reports its own win on other leaderboards.
Which reviewer wins depends on what a missed bug costs your team, because a loud, thorough reviewer overtakes a balanced one when misses cost more.
A seeded-bug test with 8 to 15 planted bugs and a short scoring script gives you your own recall and precision in about an hour.