34. In our benchmark, GPT-5 missed 40% of vulnerabilities. Opus 4.6 reduced that to 18%. GPT-5.5 brings it down furth… by @Xbow (XBOW) · backlist 2026-05-05 · rubric 92.0