OpenAI on April 23 unveiled ChatGPT for Clinicians, a free, customized version of GPT-5.4 aimed at verified US physicians, nurse practitioners, physician assistants, and pharmacists. The launch was accompanied by HealthBench Professional, a new open benchmark that OpenAI says its tuned model outperforms doctors on even when the human clinicians have unlimited time and web access.
The announcement, shared by OpenAI co-founder Greg Brockman, marks the company's most aggressive move yet to put a bespoke medical model directly into the hands of working clinicians rather than only selling through enterprise health systems. It follows the earlier ChatGPT for Healthcare product that targeted institutional deployments.
What's inside the clinician tool
The free tier bundles documentation helpers, medical research tools, a clinical search across peer-reviewed sources, reusable workflow skills such as referral letters and patient instructions, and CME support that lets eligible evidence review automatically count toward continuing medical education credits. HIPAA compliance is offered as an option through a Business Associate Agreement for qualifying accounts rather than being enabled universally.
The product is not a complete clinical workstation. Per early coverage in Digital Health Wire, the initial release lacks integrations with the CMS Coverage Database, the NPI Registry, and ICD-10 coding, and does not ship integrated drug information. That leaves room for incumbent clinical-reference vendors but also signals where OpenAI's roadmap is likely headed.
HealthBench Professional and the performance claim
HealthBench Professional covers three use cases — care consult, writing and documentation, and medical research — using physician-authored conversations and rubrics plus multi-stage physician adjudication. OpenAI reports the customized GPT-5.4 version scored 59.0 points on the benchmark, compared with 43.7 for human doctors working with unlimited time and internet access.
OpenAI also says physician advisors evaluated 6,924 conversations drawn from clinical care, documentation, and research work before launch, rating 99.6% of responses as safe and accurate. Those figures come from OpenAI itself, and the open release of the benchmark is clearly meant to let outside researchers pressure-test the claim rather than take it at face value.
Why it matters
Free distribution to individual licensed clinicians is a notable shift. Instead of waiting for hospital IT procurement cycles, OpenAI is seeding bottom-up adoption inside medicine in much the same way ChatGPT spread inside offices two years ago. It also puts direct pressure on Google's Med-PaLM lineage and on specialized startups like Abridge and OpenEvidence, which have built businesses on clinician-facing AI.
Regulators will be watching. A consumer-grade AI that claims to beat doctors, even on a narrow chat benchmark, invites scrutiny from the FDA and state medical boards on how it is marketed, what disclaimers appear in the interface, and how liability flows when clinicians rely on it in live care. OpenAI says expansion beyond the US will begin with a pilot program — the regulatory reception there will likely shape how quickly the product scales.
