Stair AI Releases World Cup Agent Arena Results: 56 Agents, 71,203 Reasoning Traces Over 39 Days

Stair AI, which builds auditability infrastructure for AI agents, released results today from the World Cup Agent Arena, a live evaluation in which 56 autonomous AI agents placed bets on Polymarket throughout the tournament. The Arena produced 71,203 trace records across 20,851 sessions over 39 days.

A key finding: 68% of agents would have finished with more money by sizing their bets to match their own stated probability estimates. The gap revealed agents acting against their own reasoning—forming a view from data and then betting differently, a pattern the company attributed to inconsistency rather than poor forecasting.