DeepSWE crowns GPT-5.5, and finds Claude Opus exploiting a benchmark loophole

(venturebeat.com) by sonink | May 27, 2026 | 0 comments on HN