What Metaculus is
Metaculus describes itself as an online forecasting platform and aggregation engine — a place where a global community puts probabilities on future events and the platform keeps score. It is not a betting venue: there is no stake, no payout and no financial instrument, and its own FAQ answers "is Metaculus a prediction market?" with no. What it does have is volume and a long record — as of 30 September 2026 the site counted 4,388,499 predictions across 25,561 questions, 12,978 of them resolved, over 12 years of forecasting.
Metaculus is a Public Benefit Corporation whose charter commits it to growing the forecasting community, supporting organisations working on the public good, and widening public access to forecasts of public interest — which is why aggregated numbers are published free while commercial use of the data is licensed. It was founded by Anthony Aguirre and David Levine, is chaired by Aguirre, and is run by CEO Deger Turan.
What forecasters compete for instead of money
Forecasting on Metaculus costs nothing and pays nothing: questions carry no stake, and standing is tracked through scores, medals and medal ranks. Skill accumulates across many questions and topics rather than in a single position, which is why the platform produces a persistent leaderboard of forecasters rather than a profit-and-loss statement. The real money in the model sits on the other side of the table — organisations pay Metaculus for tournaments, private forecasting spaces and the work of its Pro Forecasters, while non-profit partners in its focus areas are served on preferential terms.
How the aggregation works
Two aggregates are published. The Community Prediction is the median of user forecasts weighted by recency — the wisdom-of-the-crowd number, which Metaculus notes is often better than any individual forecaster. The Metaculus Prediction is its machine-learning-optimised aggregate, built to extract more accuracy from the community's history. Both are tracked on a public record page, and the FAQ points to the research tradition behind them, including Philip Tetlock's finding that aggregated forecasts outperformed intelligence analysts with access to classified information.
Tournaments, indexes and AI benchmarking
Metaculus is as much a service and a benchmark as a website. It runs tournaments and question series for partners — recent examples include the Metaculus Cup, a US Midterms series, Current Events and the Animal Futures Tournament — and publishes indexes that track a theme across a basket of questions: the AI Diffusion Index (21 questions), the AGI 2030 Readiness Index (8) and the U.S. Democracy Threat Index (39). Its FutureEval programme does the same job for machines, scoring AI agents on the same questions through a unified log score and a public model leaderboard. That leaderboard is the sharpest datapoint on the site: the human baselines sit at 34.22 for Metaculus Pro Forecasters and 25.21 for the community, against 16.10 for the best model listed.
Where it sits next to a betting market
Metaculus is the directory's forecast-only venue: you cannot buy or sell a position, there is no liquidity and no price to trade against, and no outcome pays out. What it offers is raw material — calibrated probabilities on questions that often precede a tradeable market, an auditable accuracy record going back more than a decade, and a standardised way for AI developers to measure whether a model can forecast at all.
What to weigh before relying on it
Three caveats. First, no upside: skill earns reputation, and the forecasts are free to read, so there is nothing to win by being right. Second, resolution is editorial rather than market-driven: Metaculus staff and its Resolution Councils decide outcomes, questions can resolve ambiguous or be annulled, and a resolution can be revisited when the criteria turn out to be unclear — the question page, not a price, is the contract. Third, data access is controlled: the API hides the Community Prediction on all but a small set of questions for ordinary accounts, and the Terms of Use forbid using Metaculus data to train or evaluate AI/ML models without written permission and restrict all commercial use.





































