Thoughtful launches BallotBench to test AI forecasts across 863 races
Nine agents get internet access and $500 each to forecast the 2026 midterms, with their research traces and revisions on display.
By Ryan Merket · Published
Primary source: x.com
Why it matters
BallotBench makes agent research and spending inspectable alongside their forecasts. Its results will test whether that transparency and extra research time translate into calibrated calls across hundreds of real races.
Brian Zhou, Mersad Abbasi and Karina Nguyen have launched BallotBench, a live test of nine AI agents forecasting 863 federal and statewide races in the 2026 U.S. midterms. Thoughtful Lab described the project in a thread on X published October 28th, with the run scheduled through Election Day, November 3rd.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.