Ask HN: Is Codex with GPT 5.5 Extra High being dumbed down?
A user on Hacker News reports that Codex with GPT 5.5 Extra High has noticeably degraded, comparing it to a reliable tech lead becoming an overconfident junior dev who doesn't think through problems. They question if the model is being dumbed down or quietly rerouted.
Background
- The user complains **GitHub Copilot Codex** (VS Code's AI coding assistant) and **Claude Code by Anthropic** (a terminal coding agent) have degraded over time. They mention "GPT 5.5" and "Opus 4.6-4.7" — not real public model names, but likely user labels combining rumored internal versions with effort settings (like "thinking: max" in OpenAI's o1 series).