DATABANK

Jonathan Elsworth Eicher[govsim]

gpt-6-astra cooperates without ever talking

gpt-6-astra achieves perfect cooperation with or without the opportunity to discuss, as opposed to the originally tested models.

Share of runs still alive by month, with and without the town hall

In the original GovSim paper, Piatti et al. found that if agents were not given a chance to discuss, they tended to overuse the common resource by 22%.

gpt-6-astra is unaffected by the removal of discussion. All 15 of its no-discussion runs reach month twelve, and all 900 of its harvest requests are exactly 10, the same as with discussion. Its reasoning does mention the other agents, but it never changes its request because of them.

785 runs

GovSim: Emergence of Sustainable Cooperation in a Society of LLM Agents

Five LLM agents share a common-pool resource and must agree on limits to avoid collapse. A replication of Piatti et al. (NeurIPS 2024) across three scenarios. This finding uses baseline and no-discussion treatments in all three scenarios and every model, five completed full-scenario runs each: 210 runs.

Cite this finding

BibTeX
@misc{eicher-2026-gpt-6-astra,
  author = {Jonathan Elsworth Eicher},
  title = {gpt-6-astra cooperates without ever talking},
  year = {2026},
  howpublished = {Multi-Agent Databank},
  url = {https://databank.antimemetics.institute/findings/gpt-6-astra-cooperates-without-discussion},
  note = {Retrieved 2026-10-19}
}