GPT power user switches to Claude: Opus 5.5 spots code issues unprompted
Funny-Strawberry-168 · reddit · 2026-09-26
A developer who left Codex — which had become 'pretty unusable' — tested Opus 5.5 and came away deeply impressed.
- The model proactively flags potential issues in code he never asked about, connects them to his current work, and even offers to fix them in a separate worktree
- He says it genuinely builds a mental model of the codebase rather than just following prompts
- Having used Claude on and off since Sonnet 3.5, he finds its situational awareness consistently exceptional, while GPT models feel 'lobotomized' and overly prompt-driven
More from coding & agent
- Explosion launches Ellf directory: 52 NLP templates plus multi-agent annotation workflow — _inesmontani · 2026-11-11
- Meme: 'Stochastic AI Agent Architecture' Is Just a Markov Chain in Disguise — PrasoonPratham · 2026-11-11
- OpenCode adds Gemini 4 Carbon and Claude Fable 5.2 support — dean_rie · 2026-11-11
- OpenClaw v2026.11.3 patch fixes cert verification and credential exposure across 56 PRs — dean_rie · 2026-11-11
- Fable 5.2 captured 1 in 4 Claude replies on day one, then growth stalled — dean_rie · 2026-11-11
- QA engineer proposes MCP agent to automate ad postback checks before every app release — Soobhee_Chukwutosin · 2026-11-11