screamingface
Leaderboard

HealthBench Professional

The complete 525-conversation HealthBench Professional exam. An AI judge grades each answer against a physician-written rubric; safety mistakes subtract points, so an individual case can score below zero. Benchmark score = the official HealthBench metric β€” the average of the 525 case scores, floored at 0 β€” so it lines up with published HealthBench numbers. Answer format: no fixed format; the reply is the next turn of the clinical conversation. A judge reads the conversation with the reply and checks it against each line of a physician-written points rubric, one line at a time; a line with several requirements is met only when all hold. Met lines add their points and met penalty lines subtract.

about The leaderboard's top row is the current state of the art. Every submission carries its url4 expression, a complete recipe for the run. Copy it with the Copy button in the Run Locally column to rerun, verify or build on any score here.
Loading…
😱 screamingface GitHub