FutureSearch

About FutureSearch

FutureSearch is a forecasting engine that answers questions about the future with a probability and the research behind it, rather than with a paragraph of hedged prose. You pose a question in one of six shapes — probability, numeric, date, categorical, conditional, or decision — and a team of agents searches, reads and reasons its way to an estimate you can inspect. The differentiator the company leans on is a public track record rather than a benchmark table it wrote itself: it ranks first of 197 bots on the Metaculus FutureEval live tournament, scored by Metaculus on real unresolved questions, and twentieth of 335 on the Forecasting Research Institute's contamination-free ForecastBench. It also trades real money on Kalshi, Polymarket and the S&P 500 and publishes every position with its research, losers included, which is an unusually falsifiable claim for an AI product. On its own 1,907-question pastcasting benchmark it scores 0.122 against 0.130 for the best frontier model. Beyond single forecasts there is a toolkit around them — multi-agent runs and research teams for batch work, plus an agent map for cheaper row-level passes — and the whole thing is available over an API so another agent can call it. Pricing is per question rather than per seat, which fits the intermittent way forecasting actually gets used: you pay for the questions you ask and the subscription mainly buys concurrency and a usage discount.

Key Features

FutureSearch Pros & Cons

Tags