Are you better calibrated than a frontier model?

Ten real prediction-market questions, drawn from the same 143 the frontier models answered. Give a probability, not a yes/no — you're scored on calibration, exactly like they were.

Scoring is the Brier score: (your probability − outcome)², averaged across questions. Lower is better. Saying 90% on things that don't happen is expensive; saying 55% on everything is cowardice the math also punishes. At the end you're ranked against all 15 model entries on the exact same ten questions.

Fair-play note: these markets resolved between May and July 2026 — after every model's training cutoff, but possibly not after yours. If you remember the news, you'll have an edge the models didn't. Try to answer from what you'd have known at the time. Method details on the benchmark page.

Leaderboard these ten questions only · bare condition · lower Brier is better

copied ✓

Model scores here are computed on the same ten questions you answered, from the same run as the full leaderboard — so ranks can differ from the 143-question board. "Base rate" always answers the dataset's YES frequency; "coin flip" always says 50%.