HealthBench is a new open benchmark using 5,000 physician-created rubrics on multi-turn conversations to measure LLM performance and safety across health contexts.
Conversations where the user asks for or implies they prefer a response with specifics, detailed explanations or in depth understanding
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
UNVERDICTED 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
HealthBench: Evaluating Large Language Models Towards Improved Human Health
HealthBench is a new open benchmark using 5,000 physician-created rubrics on multi-turn conversations to measure LLM performance and safety across health contexts.