OpenAI unveiled on Wednesday ChatGPT for Clinicians, a version of ChatGPT designed to support clinical tasks like documentation and medical research.
The AI company says it's making the AI tool free for any verified physician, nurse practitioner, physician assistant or pharmacist, starting in the U.S. OpenAI says it plans to expand access to additional countries and groups over time.
The company says clinician usage of ChatGPT has more than doubled over the past year. The American Medical Association reports that physicians’ use of artificial intelligence has more than doubled since 2023, with 81% surveyed by the AMA reporting that they currently use the technology in a professional context.
The new AI tool follows OpenAI's January launch of a dedicated healthcare product, ChatGPT for Healthcare, along with a consumer-focused product, ChatGPT Health, to help individuals understand their health information and get personalized answers to medical questions.
ChatGPT for Healthcare is a workspace for researchers, clinicians and administrators, powered by GPT-5 models that went through doctor-led testing. Early adopters already rolling out ChatGPT for Healthcare include Boston Children’s Hospital, Cedars-Sinai Medical Center, Stanford Medicine Children’s Health, AdventHealth, HCA Healthcare, Baylor Scott & White Health, Memorial Sloan Kettering Cancer Center and the University of California, San Francisco, OpenAI said back in January.
Enabling free access to ChatGPT for Clinicians is the next step and builds on the company's foundation of continual model evaluation for healthcare, OpenAI executives said in a blog post.
OpenAI is moving deeper into healthcare as existing clinical tools build out AI capabilities and, by contrast, AI startups in the sector develop capabilities to expand across clinical workflows. Abridge, a company that started as an AI scribe, is scaling up its clinical decision support solution through a partnership with Wolters Kluwer's UpToDate, a clinical evidence solution that's been on the market for 30 years. Wolters Kluwer also launched a generative-AI-powered version of its clinical tool, called UpToDate Expert AI, to provide rapid, evidence-based answers exclusively from the curated UpToDate database.
OpenAI also introduced HealthBench Professional, an open benchmark for evaluating large language models on clinical tasks. The benchmark is organized around three common use cases: care consult, writing and documentation and medical research. This new benchmarking tool builds on HealthBench, an open-source benchmark OpenAI released last year that's designed to measure the performance and safety of large language models in healthcare.
The new benchmarking tool uses "physician-authored conversations and rubrics, multi-stage physician adjudication, and careful data filtering to measure performance and safety in common clinician chats," according to OpenAI.
As a baseline, the company asked human physicians to produce their own responses for tasks in their specialty, with unbounded time and web access. Through HealthBench Professional evaluations, OpenAI found that GPT‑5.4 in the ChatGPT for Clinicians workspace outperforms base GPT‑5.4, all other OpenAI and external models, and human physicians, the company said.
OpenAI says it worked with "hundreds of physician advisors" to inform and improve capabilities for ChatGPT for Clinicians.
ChatGPT for Clinicians offers free access to OpenAI's current frontier models for healthcare use cases to help address questions, research and documentation. Clinicians can also turn common workflows into reusable skills so ChatGPT can follow the same steps each time for tasks like referral letters, prior auth and patient instructions, the company said in the blog post.
The tool also offers a clinical search tool that provides access to evidence from "millions of reputable, peer-reviewed medical sources," OpenAI executives said.
An OpenAI spokesperson said the models powering ChatGPT for Clinincians are trained on top publications and journals in PubMed as well as constrained web search of trusted sources like the the Food and Drug Administration, and major medical societies. Depending on the clinical context, these sources may include U.S. government agencies such as the FDA and the Centers for Disease Control and Prevention, as well as professional medical societies such as the American Hospital Association, the spokesperson said.
The spokesperson said OpenAI only uses medical sources for healthcare citations and built fit-for-purpose citation tools.
As clinicians research clinical questions in ChatGPT, eligible evidence review can automatically count toward continuing medical education credits, without separate courses or extra paperwork, according to the company.
ChatGPT for Clinicians does not include system integrations with electronic health record (EHR) systems, an OpenAI spokesperson said, noting that its enterprise offering OpenAI for Healthcare does integrate with EHRs.
Certain eligible accounts can also set up a Business Associate Agreement (BAA) with OpenAI to support HIPAA-compliant use when access to protected health information (PHI) is required. OpenAI executives noted that many clinical tasks don't require PHI.
OpenAI also asserts that conversations are not used to train models and the tool offers features like multi-factor authentication to protect the security of sensitive information.
The company claims its AI models are continuously evaluated for health performance and safety. Physician advisors have reviewed more than 700,000 model responses that reflect how clinicians and patients may use ChatGPT in the real world, the company said.
Before releasing ChatGPT for Clinicians, physician advisors tested 6,924 conversations in their daily work across clinical care, documentation and research. Doctors rated 99.6% of responses as safe and accurate, OpenAI said.
"On a subset of 355 examples where, for each, three independent physicians specified ground-truth citations, ChatGPT for Clinicians cited those sources more often than human physicians. Even so, ChatGPT for Clinicians is designed to support clinicians with information, not replace their judgment or expertise," OpenAI executives wrote in the blog post.
The company also points to third-party evaluations like Stanford’s MedHELM and MedMarks that ranked OpenAI’s models above other LLMs for real-world healthcare use.
Editor's Note: This article has been updated with additional information about ChatGPT for Clinicians' clinical information sources.