Claude Opus 5.5 Raises the Bar for AI-Assisted Interactive Design, Author Says

Summary

A Quesma blog author compares GPT-6 Astra and Claude Opus 5.5 as tools for building interactive data visualizations and explorable explanations. The author says earlier models could produce competent landing pages or useful analysis, but often required manual removal of visual clutter, overlaps, and generic “AI slop.” For a one-shot test, both models received the same prompt: create a pnpm-based Three.js visualization of all 55 imagined cities in Italo Calvino’s Invisible Cities without asking questions. GPT-6 Astra completed an end-to-end result in 53 minutes at medium effort, at an estimated API cost of about $10; the author found it capable but still marked by unnecessary concepts, comments, and accessibility-oriented text that did not belong in the visible design. Claude Opus 5.5 produced its version in 1 hour 25 minutes for about $74, using six parallel subagents whose combined agent time was estimated at roughly seven hours. The author found the result strikingly polished, while noting that it may be overly rich and insufficiently minimalist and that the system finished before the requested six-hour limit. The comparison is based on one person’s subjective experiments, not a controlled benchmark. The author expects AI to become widely used for interactive design while questioning how human creators’ roles will change.