OpenAI Launches New Clinical Chatbot
On April 22, OpenAI, the AI research and deployment company behind ChatGPT, introduced ChatGPT for Clinicians, an AI chatbot designed to support clinical tasks. According to the company, the model is available at no cost to verified physicians, nurse practitioners, physician assistants, and pharmacists in the United States.
The company reported that millions of clinicians worldwide are already using ChatGPT in clinical practice, with usage among clinicians more than doubling over the past year.
ChatGPT for Clinicians was developed with input from hundreds of physician advisors to support common clinical use cases and align with clinician workflows.
According to the announcement, health-care professionals can use the chatbot to assist with routine workflow tasks, such as drafting referral letters and prior authorizations, searching peer-reviewed medical sources to support clinical decision-making, and conducting medical literature reviews. It also highlights the ability to obtain continuing medical education credits for clinical questions within ChatGPT without additional paperwork. The company added that HIPAA-compliant use is available for eligible accounts through a Business Associate Agreement.
The model is based on ChatGPT-5.4 and includes expanded usage limits and enhanced security and privacy controls, with the promise that clinician conversations are not used to train further models. The system also provides citations from medical sources in its responses to support verification of its outputs.
Benchmarks and Evaluations
Prior to its release, OpenAI reported that physician advisors tested the model across 6,924 clinical conversations spanning care, documentation, and research tasks; the advisors rated 99.6% of responses as safe and accurate. In a subset of 355 conversations reviewed by independent physicians who provided ground-truth citations for relevant data, ChatGPT for Clinicians was more likely than physicians to cite those same sources.
OpenAI also introduced HealthBench Professional, an open benchmark designed to evaluate AI performance on real clinician tasks. It builds on HealthBench, released in May 2025, for evaluating AI-driven health-care conversations.
HealthBench Professional assesses large language model performance across three domains—clinical consultation, writing and documentation, and medical research—using structured rubrics, multistage physician adjudication, and data filtering.
In a report on HealthBench Professional, OpenAI researchers said that ChatGPT for Clinicians, based on ChatGPT-5.4, outperformed other models—including OpenAI models, Claude Opus 4.7, Gemini 3.1 Pro, Grok 4.20, and physician-written responses—on clinical tasks. The model achieved a HealthBench Professional score of 59.0, compared with 48.1 for the next-highest scoring model, ChatGPT-5.4. It also outperformed other internal and external models across all three use cases, with the largest differences observed in writing and documentation.
ChatGPT for Clinicians also demonstrated strong performance across medical specialties, achieving the highest mean score among models in 6 out of 10 evaluated specialties. In hematology/oncology, the model achieved a HealthBench Professional score of 48.6, compared with 47.4 for ChatGPT-5.4, 39.2 for Gemini 3.1 Pro, and 48.4 for physician-written responses.
Competitive Market
ChatGPT for Clinicians enters a growing field of AI chatbots for health-care professionals, including OpenEvidence and Doximity’s DoxGPT. All three platforms are currently free for verified physicians in the United States only.
Although OpenAI says that ChatGPT for Clinicians can draw on peer-reviewed studies, public health guidance, and clinical guidelines, it is unclear whether access to such medical sources is consistent across platforms.
OpenEvidence’s leadership in this area may be related to its partnerships with organizations such as the American Society of Clinical Oncology (ASCO) and the National Comprehensive Cancer Network (NCCN).
At present, the capabilities of ChatGPT for Clinicians are similar to those of OpenEvidence and DoxGPT, including documentation generation and clinical research support. Both platforms also offer enterprise-scale deployments. Preferences among these platforms may ultimately depend on additional company offerings—such as billing support and integrated dialer functionality, although their impact on adoption remains unclear—or direct comparisons.
"Fine-tuning on clinical references is a solved problem, but the real barrier to adoption is workflow friction; if an oncologist has to manually toggle between a patient record and a separate AI tab, the utility is lost. For any model to truly unseat the incumbents, it must move beyond general evidence retrieval and demonstrate it can securely handle sensitive personal health information within our daily documentation cycle," commented Ravi B. Parikh, MD, MPP, FACP, Editor-in-Chief of ASCO AI in Oncology; Associate Professor in the Department of Hematology and Medical Oncology at Emory University School of Medicine; and Medical Director of the Winship Data and Technology Applications Shared Resource at Winship Cancer Institute
Other Health-Care Offerings
In January 2026, OpenAI introduced ChatGPT Health, a consumer-facing large language model focused on health and wellness. The launch followed increasing public use of ChatGPT for questions related to personal health data. The model includes enhanced privacy, security, and data controls, enabling users to upload medical records and connect wellness apps for more personalized insights.
However, in a stress test of the model’s triage capabilities, researchers identified several issues, including missed high-risk emergencies, raising concerns about the use of AI in consumer health discussions without clinician oversight.
In a social media post, Karan Singhal, head of Health AI at OpenAI, responded to the research, noting that HealthBench uses a different approach to evaluation. He said internal testing showed stronger performance in a similar evaluation when the model was allowed to respond more naturally and ask follow-up questions. He also raised broader questions about evaluation standards for health-related AI systems, including how they should align with real-world use and how to adapt rapidly evolving models.
OpenAI has also released other health-care–focused AI products this year, including OpenAI API and ChatGPT for Healthcare, which offer HIPAA-compliant capabilities for health-care organizations’ workflows. These tools were initially rolled out to eight leading health-care institutions, although limited information has been released about their use of these platforms to date.
The company noted that ChatGPT for Clinicians is intended for individual clinicians without access to centralized AI tools within their practices or institutions, whereas ChatGPT for Healthcare is designed for health-care systems and hospitals.
ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.