What I love about GPT-5.6 Sol: it reviews like a senior engineer. It found a bug in my spec — a missing release call my code implemented faithfully — and read a dependency's pinned source to verify a threading claim. One max-effort audit: 98 files, 20 real findings.