HealthBench (incl. Hard and Consensus)

Measures: Models respond to 5,000 multi-turn health conversations with users or clinicians, graded against 48,562 conversation-specific rubric criteria written by 262 physicians; the Hard and Consensus subsets are reported separately.

Maintainer: OpenAI
Introduced: 2025
Status: Active
Saturation: Headroom
Typical provenance: Self-reported
In scoring set: No

Paper

Leaderboard

Show independent results only

Every result we hold, whoever ran it. Only independent and third-party rows count towards a score.

No results to show.