New Opus went from flawless to average, says dev, but still beats Opus 5
haider1 · x · 2026-10-04
AI coding influencer haider1 (71.9K followers) says the new Opus went from "flawless" to "average" — an emotional roller coaster — though it's still much better than Opus 5.
He also shares a practical tip: dividing work among subagents keeps the main session's context from getting bloated and results in far fewer hallucinations.
More from coding & agent
- Uncle Bob tells PR critics he'll have his AI agents handle their feedback — astralmatrix · 2026-10-04
- context-mode sandboxes AI coding agent tool output, cutting context size by ~98% — lxfater · 2026-10-04
- GitHub Copilot app adds side-by-side diff, terminal and browser panels for reviewing agent code — unixterminal · 2026-10-04
- Leave comments on Paper, agents turn them into motion code: 'design to get a better prompt' — davidhoang · 2026-10-04
- A few nights with Claude and three.js: a water sim in the browser that still wants a duck — minchoi · 2026-10-04
- "It Sends a System Message to My Agent Fleet": Reboots Now Brief AI Agents Like Engineers — kevinnbass · 2026-10-04