Y Combinator and Olam Labs test GPT-6.1 Sol and Claude Opus 5.5 agent swarms in Minecraft

Researchers at Y Combinator and Olam Labs have published results from tests in which swarms of AI agents built on OpenAI's GPT-6.1 Sol and Anthropic's Claude Opus 5.5 competed and collaborated inside Minecraft, the Minecraft scenario Bed Wars, and Mindustry, according to tomshardware.com.

In a test called Civ50, 50 GPT-6.1 Sol agents were given three hours to build a thriving civilization. The colony never thrived. The researchers noted that after the first night many ambitious initial tasks were abandoned and behavior tended toward bureaucracy. Four agents reportedly spent 11 minutes appointing a sole fertilizer operator for a single piece of bone meal on one wheat block.

In Bed Wars, 10 GPT-6.1 Sol agents faced 10 Claude Opus 5.5 agents. The GPT agents used more aggressive strategies while the Claude agents played defensively. When a tool bug stopped block-breaking, GPT agents diagnosed the problem and shared workarounds, while several Claude agents concluded that breaking was impossible.

In Mindustry, the GPT agents won 6-1 against the Claude agents across 7 games. The researchers concluded that GPT agents acted more like a hive, while Claude agents had a single leader with democratic decision-making.

The agent swarms are open sourced so others can run and test them, the researchers said.