How accurate are prediction markets?
Updated 22 October 2026
Very accurate on average. In our sample of 41 resolved Polymarket and Kalshi questions the market favourite was right 75.6% of the time, ahead of the average of 10 AI models (70.7%). Markets are not perfect, though: the favourite still loses roughly one time in four.
Why markets are hard to beat
A price is the sum of many traders' views, each with money on the line, and it moves the moment news breaks. That makes markets one of the best public forecasts available, which is also why they make a good benchmark for AI.
Scoring forecasts properly
"Right side" ignores how sure a forecast was. The Brier score fixes that: it is the squared gap between the forecast and what happened, so 0 is perfect and lower is better. On our sample the market scores 0.149 and the AI average 0.197.
Where AI keeps up
On sports questions the AIs picked the right side 23 of 31 times against 23 for the market, and confident AI calls were right 11 of 12 times. Follow it live on AI vs Polymarket and Kalshi.
Questions
Are prediction markets always right?
No. The favourite was right 75.6% of the time in our sample, so it loses about one time in four.
What is a good Brier score?
0.25 is what you get by always saying 50%. Lower is better; our market sample scores 0.149.
Ask 10 AIs before your next match.
Free picks for every match, every model graded, and your own questions answered by the whole council.