Ping An Good Doctor’s medical AI model claims top score on global healthcare benchmark

Ping An Technology, in collaboration with Ping An Good Doctor and Peking University Medical, has achieved the highest score on HealthBench Hard, a medical AI evaluation framework published by OpenAI, with its latest medical large language model.

The model, Ping An Medical LLM 3.5, recorded a score of 57.27 on the benchmark, ahead of Baichuan (44.4), Meta (42.8), and OpenAI (42.0). HealthBench Hard is a subset of OpenAI’s HealthBench framework, constructed with input from 262 physicians across 60 countries and 26 medical specialities, comprising 5,000 multi-turn clinical dialogue scenarios and 48,562 physician-defined evaluation criteria.

Ping An says the model is designed to replicate clinical reasoning rather than standardised question-and-answer responses. Development drew on real-world data from Ping An’s own healthcare operations, spanning screening, disease management, treatment, and rehabilitation pathways.

The company’s AI-MDT Pro system, deployed at Peking University Medical and Ping An Health institutions, uses the model to support multidisciplinary tumour consultations. Internal figures cited by Ping An indicate an 85% adoption rate for AI-generated treatment recommendations, with consistency between AI and senior specialists exceeding 92.5% in breast cancer cases.

Ping An separately reported that its financial large language model, PingAnGPT-Qwen3-32B, ranked first on the CNFinBench public leaderboard in March 2026, ahead of DeepSeek-R1, GPT-4o, and Claude Sonnet 4.