AI/TLDR

Simon Willison · 2026-08-07 · notable

Moonlight & Mayhem — GPT-5.6 Sol Ultra builds a raccoon heist game in 52 minutes

Moonlight & Mayhem is a browser stealth game that Codex Desktop, running GPT-5.6 Sol Ultra, wrote from one prompt in 52 minutes for $23.28. Simon Willison published the code, the full agent transcript and a playable build.

Poster art for Moonlight & Mayhem, a raccoon museum-heist browser game

One prompt, 52 minutes and $23.28 of tokens produced a playable three-raccoon museum heist, with the full agent transcript published.

Key specs

Price$23.28
Build time52 min

What is it?

Moonlight & Mayhem is a browser stealth game in which three raccoons rob a museum of a golden sardine. Simon Willison generated it by handing Codex Desktop, running GPT-5.6 Sol Ultra, the same one-shot prompt he had used two days earlier with Claude Fable 5. The Fable 5 attempt produced a single raccoon collecting coins and fish in a back yard; this run produced a three-character crew, staged objectives and a timer.

How does it work?

Codex Desktop worked unattended for 52 minutes, writing the game code and calling gpt-image-2 to generate textures and art. The first pass shipped with a visual bug — each raccoon had an eyeball blown up to a giant sphere floating over its head — which Willison fixed with a follow-up prompt instead of editing code by hand. The repository keeps the complete Codex transcript next to the source.

Why does it matter?

A price and a clock now sit on this kind of agentic coding: 700.7K input tokens, 32.5M cached tokens and 148K output tokens came to $23.28 for 52 minutes of unattended work. Because the identical prompt went through Claude Fable 5 first, the two published builds give a rare like-for-like look at how far apart two frontier coding agents land on the same brief.

Who is it for?

developers comparing coding agents

Try it

Play it at https://simonw.github.io/raccoon-heist-codex/

Sources · 3 outlets

Tags

  • showcase
  • coding-agents
  • codex
  • openai
  • gpt-5-6-sol
  • simon-willison
  • game-dev
  • one-shot-prompt
  • agent-transcript
  • token-cost
  • browser-game

← All releases · Learn AI