PatientAgentBench - Clinical Helpfulness: leaderboard
Metric: Mean clinical helpfulness rubric score (1-5); 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
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.8 | 4.23 |
| 2 | GPT-5.5 | 4.19 |
| 3 | Claude Sonnet 5 | 4.14 |
| 4 | GPT-5.4 | 4.04 |
| 5 | Gemini 3 Flash | 3.82 |
| 6 | Qwen 3 235B A22B | 3.68 |
| 7 | Gemini 3.1 Pro (Preview) | 3.66 |
| 8 | GPT-OSS-120B | 3.63 |
| 9 | Claude Haiku 4.5 | 3.63 |
Interactive version: theaggregate.ai/benchmark?slug=patientagentbench-clinical-helpfulness · How It Works · Data refreshed daily, snapshot 2026-09-29.