█

AI/TLDR

Quesma · 2026-10-07 · notable

Invisible Cities in 3D — Claude Opus 5.5 and GPT-6 Astra get one prompt

Piotr Migdał gave Claude Opus 5.5 and GPT-6 Astra the same one-line prompt: visualize all of Calvino's Invisible Cities in three.js. Astra took 53 minutes for about $10; Opus took 85 minutes for about $74.

Quesma blog thumbnail for the Invisible Cities one-shot visualization test

One prompt, two coding agents, two live 3D versions of Italo Calvino's Invisible Cities.

What is it?

Piotr Migdał of Quesma ran a one-shot test: the same prompt asking for a three.js visualization of every city in Italo Calvino's Invisible Cities, given to GPT-6 Astra in Codex and Claude Opus 5.5 in Claude Code, with up to six hours allowed. Both results are published as live demos with their source code.

How does it work?

GPT-6 Astra worked for 53 minutes at medium effort and cost about $10 in API tokens. Claude Opus 5.5 ran six subagents in parallel for 1 hour 25 minutes and cost about $74. Opus said it had used roughly half of the six hours, even though it stopped well before that.

Why does it matter?

The comparison shows the price-quality trade-off between today's top coding agents on an open-ended creative task. Migdał calls the Opus result a leap in quality and the Astra one full of AI design slop, but it is a single run judged by one person.

Who is it for?

developers choosing a coding agent

Try it

https://p.migdal.pl/invisible-cities-opus-5.5/

Sources · 4 outlets

Tags

  • claude-opus-5-5
  • gpt-6-astra
  • claude-code
  • codex
  • three-js
  • one-shot
  • coding-agents
  • model-comparison
  • quesma

← All releases · Learn AI