Predicting the future is a growth business. FutureSearch, the AI forecasting platform founded in August 2023, has completed its transition from public beta to a full public launch, and it is arriving with receipts.
The company says its AI outperforms the median human superforecaster, a claim it backs with benchmark results from two of the most credible forecasting platforms in existence: Metaculus and ForecastBench.
What FutureSearch actually does
The platform takes questions, ranging from probability and numeric estimates to date and categorical forecasts, and generates structured predictions by running optimized web research agents across available information.
The technical backbone includes what the company calls shared world modeling, paired with ReAct-style research agents designed to gather and synthesize real-time context before producing a forecast.
Founder Dan Schwarz brings specific credentials to this problem. He previously served as CTO at Metaculus, the prediction market platform, and worked on Google’s internal prediction market before that. Co-founder Lawrence Phillips rounds out the leadership team.
The benchmark numbers are hard to ignore
In July 2026, FutureSearch ranked first out of 166 AI systems in Metaculus’s Summer FutureEval tournament. The platform also surpassed the superforecaster median on ForecastBench during separate evaluations. Superforecasters are the roughly top 2% of human forecasters identified through Good Judgment Project research, selected for consistently outperforming prediction markets and statistical baselines.
Scott Alexander highlighted FutureSearch in July 2026 as the only currently available AI superforecaster.
FutureSearch says it delivered a 91% annualized return after fees on a real-money portfolio traded through Kalshi, the regulated prediction market exchange. That figure is unaudited by any third party mentioned in available sources. Kalshi operates regulated event contracts in the United States, which means the trades are real, the money is real, and the outcomes are public.
Growth is accelerating, not slowing
By early August 2026, FutureSearch had crossed 3,500 registered users. More striking: 2,400 of those sign-ups happened in the 30 days prior to that count.
The platform offers both API access and research plans, positioning it for institutional and developer integrations alongside individual use.
FutureSearch has publicly committed to maintaining transparent track records. Most AI tools make vague capability claims. Publishing calibrated forecasting records exposes the model to exactly the kind of falsification that reveals whether the claims are real.
What this means for investors and markets
FutureSearch is not a crypto company. It does not use blockchain, does not issue tokens, and has not announced any integration with decentralized finance protocols.
The broader implication is competitive. If AI forecasting systems can consistently outperform human judgment on calibrated probability questions, the value of human analyst opinion in policy, risk management, and capital allocation starts to compress. FutureSearch is emphasizing use cases in AI safety and policy alongside finance.