← All benchmarks

PhysicianBench

Professional Work · 2026-05-04

Stanford HealthRex Lab's 100 physician-validated agent tasks adapted from real primary-care-to-specialist consultations across 21 specialties, each run in its own HAPI FHIR R4 electronic health record through the publisher's minimal tool-calling agent. An agent must retrieve patient data, reason clinically, place FHIR orders and write documentation, and a task passes only when all of its 5-9 checkpoints pass. Pass@1 is the mean share of tasks completed over three independent runs.

Top models (higher is better)