OpenAI gives U.S. clinicians free ChatGPT and a harder HealthBench

What the tweet revealed

Karan Singhal, who works on health AI and safety at OpenAI, described the launch in two bullets: ChatGPT for Clinicians, a free version of ChatGPT designed for clinical work; HealthBench Professional, a new benchmark to evaluate real clinician chat tasks.

His account usually posts health-AI research, model evaluation, and OpenAI health product notes. The tweet is material because it joins two things that should not be separated in clinical AI: a product surface for real users and a benchmark designed around clinician-style tasks.

Context from OpenAI’s linked rollout

OpenAI’s article says ChatGPT for Clinicians is a free version for verified U.S. clinicians, including physicians, nurse practitioners, physician assistants, and pharmacists. The company frames the product around administrative and clinical-support workflows, not autonomous diagnosis. That boundary matters: healthcare users need documentation help, chart review, patient communication drafts, and literature synthesis, but liability and local policy still constrain final decisions.

The article includes several concrete evaluation claims. OpenAI says physician advisors reviewed 6,924 conversations and rated responses as safe and accurate 99.6% of the time. It also points to HealthBench Professional, a harder benchmark for real clinician chat tasks. OpenAI separately notes that physician AI use is rising, citing 72% in 2024 versus 48% in the prior year.

The watch item is not simply adoption. It is whether the benchmark tracks the edge cases clinicians actually care about: ambiguous symptoms, medication interactions, incomplete charts, and instructions that must be adapted to local protocols. Regulators and hospital systems will also look for audit logs, data handling, and clear boundaries around patient-specific advice. A free product can spread quickly; durable trust will depend on reproducible safety evidence outside OpenAI’s own review process.

Sources: X source tweet · linked source