Sonnet build, Opus fix

One picture, four 3D worlds. Sonnet builds, Opus fixes.

Claude Sonnet 5.5 in Claude Code built one AI-generated picture as a Three.js world on low and on medium effort. Then a fresh Claude Opus 5.5 session on xhigh improved a copy of each build. One run each, with no human edits.

This is not a CmdBench benchmark run: no checks, scores or verdicts.

AI-generated picture of a miniature island: a wooden cabin with a stone chimney, a jetty and rowboat, a yellow crane over a half-built timber frame, lumber stacks, rocks and pine trees, in warm late-afternoon light.
Reference picture, AI-generated. Every run got this picture and the same prompt.

The four worlds

One run each. A fix is a fresh session that got a copy of its base build, FIX.md and one screenshot of the base's opening view. Each world is served exactly as the agent wrote it, with no human edits, and opens full screen in this tab.

Costs are Claude Code's own estimates at API list prices.

How it was made

Picture
AI-generated. The exact PNG every run got
Prompt
The same for every build: read the prompt (TASK.md)
Fix prompt
The same for every fix, with one screenshot of its base: read the fix prompt (FIX.md)
Models
Claude Sonnet 5.5 (claude-sonnet-5-5), Claude Opus 5.5 (claude-opus-5-5)
Agent
Claude Code 2.1.319
Effort levels
Claude Sonnet 5.5 on low and medium, Claude Opus 5.5 on xhigh for the fixes; one run each
Tools
File tools only: no browser, no shell, no web. No run saw its own world running; a fix saw one screenshot of its base.
Time limit
60 minutes per run
Three.js
0.187.0
Human edits
None. Each world is served exactly as the agent wrote it.
Run date