From Wikipedia, the free encyclopedia

Infographic showing AI-agent groups splitting into smaller groups and more capable models coordinating more than 1,000 agents.
Large AI-agent groups may split until they are small enough to reach consensus, while more capable language models can maintain coordination in groups of 1,000 or more. (Created by Giordano De Marzo, with assistance from AI/Claude Design, Anthropic. Based on data from De Marzo et al.,Sci. Adv., 2026)

Two graphs comparing AI-agent coordination time and maximum coordinated group size across language models.
Coordination time increases with group size, while the maximum group size capable of reaching consensus generally rises with a model’s language capabilities. Claude 3.5 Sonnet remained coordinated at 1,000 agents, the largest group tested (De Marzo et al.,Sci. Adv., 2026).

Kaynak: ScienceAlert