OpenAI released ChatGPT for Clinicians on 22 April, a purpose-built version of ChatGPT designed to support clinical workflows including documentation, medical research, and patient communication. The tool is free for any verified physician, nurse practitioner, physician assistant, or pharmacist in the United States, with international expansion planned through the Better Evidence Network. It provides access to OpenAI's current frontier models — including GPT-5.4 — and includes a clinical search tool that draws from millions of peer-reviewed medical sources with full citations, deep research capabilities for literature review, and reusable 'Skills' templates for recurring workflows such as referral letters, prior authorisations, and patient instructions. Clinicians can also earn Continuing Medical Education credits for research conducted within the platform.
The safety validation is substantial. Before launch, physician advisors tested 6,924 conversations across everyday clinical work — care consultations, documentation, and research — with 99.6 per cent of responses rated safe and accurate. Over 700,000 model responses underwent physician review during the development process. OpenAI also released HealthBench Professional alongside the launch, an open benchmark for real clinician chat tasks. On this benchmark, a custom GPT-5.4 configuration scored 59.0 points versus physician-written responses at 43.7 points — with the physicians given unlimited time and full internet access. The base GPT-5.4 scored 48.1, while competitor models ranged from 36.1 to 47.0. Conversations are excluded from model training by default, and optional HIPAA-compliant Business Associate Agreements are available for handling protected health information.
For the COR Summit community, this release signals a broader trend in how AI companies are verticialising their products. Rather than building one general-purpose chatbot, OpenAI is creating domain-specific versions with specialised tooling, safety validation, and compliance frameworks tailored to high-stakes professional environments. The clinical search with citations, the reusable Skills system, and the CME integration represent the kind of domain-specific infrastructure that makes frontier models genuinely useful in professional practice rather than a novelty. The HealthBench Professional benchmark also sets a new standard for evaluating AI in clinical contexts — moving beyond academic benchmarks to real-world physician workflows. For developers building AI-assisted professional tools, the architecture here — verified identity, domain-specific search, reusable templates, compliance options — is a blueprint worth studying.
