Hey Leynos, we used Claude Sonnet 4.5 and benchmarks we used were the Martian code review bench: https://codereview.withmartian.com/?mode=offline