PhysicianBench: Evaluating LLM Agents in Real-World EHR Environments

PhysicianBench is a new benchmark built from 100 real physician tasks in electronic health record environments.

Researchers evaluated 13 LLM agents on these execution-verified clinical workflows.

The agents struggled most on long-horizon, multi-step tasks that mirror real physician work.

The results show a clear gap between current AI agents and the demands of verified clinical systems.