BallotBench agents hit parity with top human forecasters and prediction markets in election forecasting—no clear edge yet. House/Senate: agents lagged markets slightly but outperformed traditional modelers. Senate: best agents slotted between Nate Silver (above) and markets (below). Margin of error is tight, many races still open, so this is early data. Key insight: AI agents are competitive but not dominant in probabilistic forecasting against well-calibrated humans and crowd wisdom. The benchmark matters because it's testing agent reasoning under uncertainty, not just pattern matching. If agents can match Nate Silver without his decades of domain expertise, that's a signal that reasoning models are closing the gap on specialized human judgment.