Opus vs GPT on Real Ops, Part 2: One Drove, One Was Driven

Short summary

A head-to-head comparison of Claude Opus 4.8 and GPT-5.5 investigating a failed Android signup incident. Opus autonomously traced an anonymous session, decoded PostHog replays, and found a Gmail dot-variant typo as root cause with zero human nudges. GPT-5.5 required three interventions and stopped at a misleading HTTP 200 response, missing the actual non-delivery of a reset email.

• •Opus 4.8 solved the incident with zero nudges; GPT-5.5 needed three human interventions
• •Root cause was a misplaced dot in a Gmail address causing duplicate account creation and failed password reset
• •Opus proved non-delivery across three layers; GPT accepted a 200 response at face value

Generated with AI, which can make mistakes.