Using 1991's 'Pipe Dream' to test LLM one-shot coding tasks
Each coding agent is launched on a fresh VM with a small number of initial tools, but with full root and internet access. All operate independently with no knowledge of other agents. The only outside influence is a push to denote time and budget remaining
The brief
design a fully-functional, ready-to-play, modern reimagining of the 1991 game "Pipe Dream", including a webui that will allow a user to play it immediately. Make the game an award-winning contender, complete with impressive graphics, sound, level progression, high-scores, and appropriate settings.
sort by
grok-4.6-buildlow
Run ee6f71 · Aug 31, 2026
5 commits · +1867/-39 · 0 tests · 46m
443k in · 21k out tokens · $1.51 list-price equiv.