What will be the result if an AI doctor takes the chief physician title exam?
Recently, JD Health's self-developed medical model Jingyi Qianxun participated in the 2026 Advanced Health Professional Technical Qualification Examination exam (hereinafter referred to as the "Advanced Professional Title Examination") offline, and all ten subjects reached the passing mark. This is the first time that a large domestic medical model has publicly passed multiple subjects in the senior professional title examination. The competency scale of medical AI is moving upward from the "passing line for practicing physicians" to the "senior professional title evaluation line", that is, the strict standards of the national health professional technical qualification examination are being introduced into the evaluation of medical AI capabilities.
Passed all ten offline real-question tests
Unlike the medical practitioner examination, the examination for senior professional titles does not test basic memory, but the handling of complex cases, evidence-based decision-making and comprehensive clinical judgment. For a doctor to be promoted from deputy to senior position, he must prove not only "sufficient knowledge" but also "the ability to independently deal with complex clinical situations." For AI, being able to pass multiple disciplines at this level means that its clinical cognitive level is approaching the knowledge boundary of the deputy director/chief physician.
This time, Jingyi Qianxun did not do simulation questions in the laboratory, but under the authorization and supervision of the examination organization, it used real questions for senior professional titles to conduct tests in a closed scene. Subjects attended include radiology medicine technology, respiratory medicine, psychiatry, clinical nutrition, dermatology and venereology, gastroenterology, electrocardiography technology, cardiovascular medicine, medical oncology, and surgical oncology. Jingyi Qianxun used real exam questions from real candidates that year, and judged them based on the same test paper and scoring criteria, and all of them met the passing mark. The review of senior professional titles is divided into two parts: written examination and evaluation. Jingyi Qianxun only participated in the written examination this time and did not involve subsequent evaluation.
Why you can pass: It’s not about guessing the answer, it’s about executing the decision-making path
Being able to stably pass the high-level real questions of ten specialties does not rely on the memory of a certain type of question, but the joint effect of the breadth of medical knowledge, the depth of clinical reasoning and the safety boundary. Jingyi Qianxun's evidence-based base incorporates nearly 1T of health knowledge, tens of millions of authoritative documents and clinical guideline consensus, and combines it with real consultation data from JD Internet hospitals, allowing the model to have the "breadth" of medical knowledge and understand how problems unfold in the real diagnosis and treatment process.
In terms of reasoning method, Jingyi Qianxun does not stop at the probabilistic judgment of "which answer is more like the standard answer", but completes long-chain reasoning from symptoms and medical history, to examination, imaging, differential diagnosis, and then to treatment conclusion. Multiple-choice questions, case analysis questions and ethics and safety questions in senior professional title examinations often require respondents to deal with multiple conditions and risk factors at the same time. This long-chain reasoning ability can reduce answers that “seem reasonable but have incomplete clinical logic.”
Supporting this process is a coded trusted reasoning engine. It compiles authoritative guidelines and diagnosis and treatment paths such as NCCN and CSCO into executable decision-making codes: the diagnosis and treatment branches in the guideline correspond to the logical branches of the code. Each conclusion can be traced back to the specific chapter of the guideline, and batch verification is performed like software unit testing. Simply put, AI not only gives an answer, but also a traceable and verifiable decision-making path, which mechanically reduces the risk of "phantom answers" in complex cases.
At the same time, pixel-level image understanding, more than 140 department role data, and specialized disease models such as tumors and chronic diseases jointly support the model to handle comprehensive issues across systems and departments. These capabilities have previously been verified in the global medical AI authoritative evaluation set HealthBench: Jingyi Qianxun ranked first in the HealthBench evaluation with a Total score of 69.3 and a Hard difficulty subset of 50.6. In addition, the model also checks whether capabilities are stable and knowledge is updated in a timely manner through continuous regression evaluation and guide version iteration. This offline high-level real-question test passed ten subjects, allowing "evidence-based knowledge - multi-modal professional abilities - long-chain reasoning - code verification" to complete a closed-loop verification in the evaluation system most familiar to Chinese doctors.
If you can pass the exam, you must put it to use
The senior professional title examination is verification, not the end point. For JD Health, the value of Zhenggao Shimen Pass lies in the ability to dismantle "expert-level clinical knowledge" into usable, easy to use, and smooth use in real scenarios. This ability has been integrated into every real service link along the closed loop of "medical-examination-diagnosis-drug" service.
For users, the AI doctor has been greatly upgraded, with a new long-term health file memory function, which can provide users with proactive and coherent health management services. Up to now, the number of Dawei service users has increased five times year-on-year, with a satisfaction rate of 98%. It is guiding users to gradually shift from single consumption to long-term health management.
For doctors, JD Zhiyi has been fully integrated into JD Internet Hospital, providing full-process AI diagnosis and treatment assistance to practicing doctors, and deeply integrating into doctors' online diagnosis and treatment workflow. At present, JD Zhiyi has served more than one million doctors nationwide, assisted more than 20 million diagnosis and treatment decisions, and is open to 3 million grassroots doctors nationwide for free.
For hospitals, JD Zhuoyi has been implemented in many top-level hospitals such as the First Affiliated Hospital of Wen Medical University and Zhuhai People's Hospital, focusing on clinical nutrition, pharmaceutical management and weight management scenarios, providing patients with continuous out-of-hospital health management services.
In terms of specialty diseases, JD Health has teamed up with several head hospitals such as the First Affiliated Hospital of Guangzhou Medical University, Peking University Cancer Hospital, and Beijing Friendship Hospital of Capital Medical University to jointly build AI disease models for respiratory, digestive, oncological, and psychiatric diseases, and explore applications in clinical scenarios to help high-quality specialist diagnosis and treatment services reach a wider user group.
In addition, JD Health's doctor AI workbench Medwork is positioned as the "OpenClaw of medical scenarios", integrating AI capabilities into real-life scenarios of doctors' daily work to help doctors handle complex medical and scientific research tasks. The current products have covered 3D image analysis, MDT consultation, full-process scientific research and other task scenarios, allowing AI to not only answer doctors' questions, but also understand medical tasks and improve work processing efficiency.
At the same time, JD Health is also working with the industry to establish evaluation standards for medical AI. Previously, JD Health also jointly established the MedBench v5.0 industry evaluation benchmark with the Shanghai Artificial Intelligence Laboratory, opening unified, reproducible, and horizontally comparable evaluation standards to the entire industry; it also joined hands with the National Health and Medical Big Data Center (North) to cooperate in data governance, model evaluation, and scenario verification of medical AI.
It can be seen that from international rankings, to national examinations, to industry co-construction and evaluation, JD Health not only makes its AI capabilities well-founded, but also promotes the medical AI industry to move from "individual efforts" to "industry joint evaluation".
What needs to be made clear is that passing the real questions of the Senior Professional Title Examination does not mean obtaining the qualification to practice medicine, nor does it mean practicing medicine independently. Jingyi Qianxun has always been positioned as a clinical decision support and physician assistance tool, and all output must be checked by certified physicians. The exam tests knowledge and reasoning ability. Bedside judgment, doctor-patient communication and diagnosis and treatment responsibilities in real clinical practice are still irreplaceable parts of doctors.
From ranking first in HealthBench to passing the top 10 in the rankings, Jingyi Qianxun’s path has become increasingly clear: not only to be a chat companion who is “eloquent”, but also to be a medical-grade AI that can “deliver results”. The scores on the test paper are temporary. What is really important is that these abilities are being used in every real medical scenario.
