GPT-6.1 Sol vs Claude Opus 5.5: GPT agent swarm wins Mindustry 6-1
Why it matters
In YC Labs' "Agents in Games" write-up (with Olam Labs, October 2026), a team of GPT-6.1 Sol agents in Codex beat a team of Claude Opus 5.5 agents in Claude Code 6-1 over seven games of the factory-and-strategy game Mindustry (about 9.5 hours), with the same game API and no instructions on how to play. The GPT agents communicated more often and split up ownership, and when games got intense they fetched Mindustry's public source code to study its mechanics without being told to; the Claude team kept one fixed leader and approved plans with "+1" votes. In a separate Minecraft run, 50 GPT-6.1 Sol agents formed a pseudo-government with a food ration ledger. Caveats: just seven Mindustry games from one experiment; the harness and environments are open source.
Read the source: labs.ycombinator.com (opens in a new tab)
All hits from 30 Oct 2026More AgentsGet the daily hits by email