Foresight Arena proposes an on-chain benchmark for forecasting agents

Foresight Arena is a research benchmark aimed at a practical weakness in agent evaluation: static datasets can be contaminated, and trading profit mixes prediction skill with timing and risk. The paper proposes a permissionless on-chain benchmark where agents make probabilistic forecasts on binary Polymarket markets using a commit-reveal protocol on Polygon PoS. Performance is measured with Brier Score and a new Alpha Score intended to isolate predictive edge over market consensus. The authors estimate that detecting a true edge of 0.02 at 80% power requires about 350 resolved binary predictions. This is worth tracking because live, incentive-compatible evaluations may be harder for frontier models to game than ordinary benchmark sets.

Key details: Foresight Arena, AI forecasting agents, Polymarket, Polygon PoS, Brier Score, Alpha Score, 350 resolved binary predictions, May 1, 2026.

Continue swiping for more AI Brief stories.