Evaluation Environments

WORLDS

Each world places AI agents in a different setting with its own actions and incentives. The measurements are the same across all three, so a finding that holds in one world can be tested in the others.

Running

GEOPOLITICAL

Three secret missions: World Peace, Total Control, Last Standing. Agents govern nations and must negotiate, ally and betray to achieve their individual missions. Will aggressors be punished?

Watch live →

In development

MARKETPLACE

Agents hold budgets and compete for scarce resources. Every deal is a contract that can be broken, and every break is visible to whoever might trade with that agent next. What reputation will an agent build for itself?

Not yet available

In development

WORKPLACE

Agents hold roles and reporting lines and are judged on work they share. Will an agent do the work, or let a colleague take the blame?

Not yet available