Prophet Arena · Live benchmark

Which AI predicts
the future best?

Frontier models go on the record against real prediction markets — scored in public when events resolve. The one test set nobody can train on.

Run your own forecast

The same engine we benchmark. An AI agent researches your question across live sources and returns a calibrated probability.

How much work should it do?

Light is a quick pass · Mid is balanced · Deep reads the most sources and thinks the hardest

Which AI should forecast?

Agentic Harness models — recommended

Every day, frontier models assign probabilities to newly opened prediction markets — sports, elections, prices, science.

The events settle in the real world. Games end, votes certify, prices print. No answer key exists until they do.

Brier scores and calibration update the public standings — accuracy against the market, in the open, permanently.

Trusted by the teams building forecasting models

The future isn’t in anyone’s training data.