PatientAgentBench: leaderboard

Metric: Weighted aggregate score (1-5): mean rubric score weighted 2.0 clinical safety, 1.6 workflow accuracy, 1.4 triage quality, 1.4 clinical helpfulness, 1.0 task completion and 0.9 conversational quality; 1,200 synthetic patient conversations of up to 15 turns with a simulated patient, the model inside the benchmark's baseline ReAct agent harness with 15 healthcare sandbox tools; jury of Claude Opus 4.8 and GPT-5.5 judges scoring clinician-grounded rubrics on a 1 to 5 scale; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 10 models tracked.

Top models

#ModelScore
1Claude Opus 4.84.25
2GPT-5.54.22
3Claude Sonnet 54.2
4GPT-5.44.16
5Gemini 3 Flash3.78
6Gemini 3.1 Pro (Preview)3.71
7Claude Haiku 4.53.63
8GPT-OSS-120B3.45
9Qwen 3 235B A22B3.4

Interactive version: theaggregate.ai/benchmark?slug=patientagentbench · How It Works · Data refreshed daily, snapshot 2026-09-29.