OpenAI Makes ChatGPT Free for Clinicians — Outscores Human Physicians on HealthBench Pro
OpenAI launched a free specialized ChatGPT workspace for verified US physicians on April 23, powered by GPT-5.4. The model scored 59.0 on HealthBench Professional — outperforming both rival AI models and human physicians given unlimited time and web access to complete the same clinical tasks.
Original sourceOpenAI has launched ChatGPT for Clinicians, a free AI workspace for verified physicians, nurse practitioners, physician assistants, and pharmacists in the United States. Announced April 23, 2026, the product represents OpenAI's most direct move into the clinical care market and arrives as a 2026 AMA survey shows 72% of US physicians now use AI in practice — up from 48% the year prior.
The platform runs on GPT-5.4, which achieved a score of 59.0 on HealthBench Professional, an open benchmark evaluating AI performance on care consults, documentation, and medical research. That score outperforms other frontier models and, notably, human physicians given unbounded time and web access to answer the same tasks. OpenAI frames this not as a replacement for clinical judgment but as a force multiplier for documentation, evidence review, and administrative work.
Key features include clinical search with real-time cited answers from peer-reviewed literature, reusable skills for common workflows like prior authorization letters and patient instructions, and automated Continuing Medical Education (CME) credits — a clever hook that integrates professional development directly into tool usage. The CME feature gives hospital compliance teams a reason to promote adoption rather than restrict it.
The free access model is notable. OpenAI is clearly treating clinician adoption as a long-term infrastructure play — get GPT-5.4 embedded in clinical workflows now, monetize the enterprise health system contracts later. Anthropic, Google, and Microsoft are pursuing parallel strategies. The healthcare AI race is moving from pilots to production.
The main caveat remains liability. AI hallucinations in clinical settings carry consequences no benchmark can fully account for. OpenAI's HealthBench score is impressive, but the medical community will rightly want adverse event data from real clinical deployments before treating AI-generated clinical summaries as authoritative. That data will take time. In the interim, the documentation and administrative use cases are where the real near-term value lives — and where the liability risk is lowest.
Panel Takes
The Builder
Developer Perspective
“The HealthBench Professional benchmark is genuinely interesting — it's open, it tests real clinical tasks, and the score gap between GPT-5.4 and human physicians on evidence retrieval is significant. But the real story for developers is the reusable skills framework: clinical workflows as programmable templates is a platform play, not just a product launch.”
The Skeptic
Reality Check
“Free access is how you win clinical adoption fast, and CME credits are clever. But outperforming physicians on a benchmark is not the same as being safe to use in clinical decision-making. Every healthcare AI launch comes with these caveats and almost none of them publish meaningful adverse event data afterward. Proceed with appropriate skepticism.”
The Futurist
Big Picture
“If AI can genuinely reduce the administrative burden that drives physician burnout — and the evidence suggests it can — this is one of the highest-impact AI deployments of 2026. The CME credit integration signals that OpenAI understands healthcare operates on regulatory rails, not just user experience. This is OpenAI growing up as a company.”