Olam Labs researchers put 50 GPT-6.1 Sol agents in a Minecraft survival world for three hours, asking them to build a thriving civilization. After 22 agents died on the first night, the survivors stopped exploring and building and organized ration lists, fishing shifts, and a written protocol for handing off items.
In separate Bed Wars and Mindustry matches, GPT teams beat Claude Opus 5.5 teams 6-3 and 6-1, respectively. The researchers allowed agents to access game code but not modify the running games. Olam Labs released the Mindustry environment and its Multi-Agent Harbor framework as open source and announced plans to add Diplomacy to the arena.
AI 根据报道生成 · 5 小时前更新