Guide · scoring yes/no forecasts

What is a good
Brier score?

0 is perfect. 0.25 is a coin flip. Move the slider to see why.

Your score

0.090

Always 50%

0.250

Perfect

0.000

(0.70 − 1)² = 0.090 · lower is better · 0 is perfect, 1 is the worst possible

Real scores on outcomes the model never saw

Prediction markets: Brier 0.126 against 0.241 for always guessing the base rate. ForecastBench questions: 0.185 against 0.234.

0%0%50%50%100%100%

Dots = forecasts grouped by stated chance (size = how many) · dashed line = perfectly calibrated

Zeno replay · held-out confirmation period, Jun–Sep 2026 · earlier research model v0.89.4, not the model serving today · n 1,187 market and 2,592 ForecastBench outcomes

evidence +
Uncertainty · 95% 0.117–0.135 (markets) · 0.178–0.192 (ForecastBench)Run files on Hugging Face ↗

What is the Brier score?

The average of (forecast − outcome)², where the forecast is a probability from 0 to 1 and the outcome is 1 if it happened, 0 if not. Lower is better.

How do you calculate it?

For each question, subtract the outcome from your probability and square it. Average across questions. A 70% forecast that came true scores (0.7 − 1)² = 0.09.

What is a good Brier score?

0 is perfect and 0.25 is what always saying 50% earns. A good score depends on the questions: compare against a simple reference on the same questions, such as always guessing the base rate. Beating that reference is what counts.

How do you interpret it?

Read it with a calibration chart: do 70% calls come true about 70% of the time? A low score with poor calibration usually means the questions were easy, not that the forecaster was reliable.

Score your own forecast →See every field's results →