Today we're releasing SWE-check, a specialized bug detection model we RL-trained with
@appliedcompute that matches frontier performance on internal in-distribution evals and makes meaningful progress on out-of-distribution evals, all while running 10x faster.