According to @claudeai, creators used Claude Opus 5 to build a 25-room companion website, showcasing multi-agent chat and scalable AI companion use cases.

According to @bcherny, Claude models now resist prompt injection, matching OpenAI’s new model and Gemini Flash, improving agent security across businesses.

According to @claudeai, biology fallbacks dropped ~85%, expanding safe help for health and education while dual‑use queries still route to Opus 5.

According to @galnagli, Opus 5 excels at web hacking but struggles at coding, surfacing critical vulns and noisy code output, per posts on X.

According to Ethan Mollick, Opus 5 outperforms 4.8 but inherits Fable-like dense prose, showcased in a Middle Earth railroad game demo.