FutureSearch just went live for real. The AI forecasting platform, which started out in August 2023, flipped the switch from public beta to full launch this month and showed up with hard numbers to back its claims. The company says its system outperforms human superforecasters, the roughly top 2% of prediction market participants who consistently beat markets and statistical models.
The proof sits in benchmarks from Metaculus and ForecastBench. In July 2026, FutureSearch ranked first out of 166 AI systems in Metaculus's Summer FutureEval tournament. It also surpassed the superforecaster median on ForecastBench in separate testing. Scott Alexander called it out as the only AI superforecaster currently available to the public.
What the numbers actually show
FutureSearch delivered a 91% annualized return after fees on a real-money portfolio traded through Kalshi, the regulated prediction market exchange. The trades are live, the money is real, and the results are public record. That unaudited figure still stands out in a market where most forecasting systems struggle to beat basic statistical baselines.
How does it work? The platform takes questions ranging from probability estimates to categorical and date forecasts, then spins up research agents to crawl the web for context. It calls the approach shared world modeling, paired with ReAct-style agents that gather and synthesize real-time information before producing a forecast. The technical approach matters because it means FutureSearch isn't just running pre-trained models, it's actively researching live conditions.
The founder isn't some random AI enthusiast either. Dan Schwarz was CTO at Metaculus before this, and before that worked on Google's internal prediction market. His co-founder Lawrence Phillips rounds out the core team. That pedigree runs through the entire operation.
Adoption is moving fast
By early August 2026, FutureSearch had already crossed 3,500 registered users. The growth curve is steepening, not flattening. User retention numbers and engagement metrics suggest the platform is hitting product-market fit faster than typical forecasting tools. Real people are actually using this thing for real money, not just kicking tires in a sandbox.
The timing matters too. Prediction markets and AI forecasting have gone from obscure to attracting serious capital and attention. Kalshi just got regulatory approval to expand. The infrastructure finally exists to turn forecasting accuracy into actual returns.
This article is informational. FutureSearch's results are backtested and claimed, not audited by third parties. Always do your own research before making any financial decisions or trades based on forecasting tools.
