OpenAI Debuts ChatGPT for Clinicians Free Access

OpenAI launched ChatGPT for Clinicians, a free, verified-access version of ChatGPT for U.S. physicians, nurse practitioners, physician assistants, and pharmacists, designed to accelerate documentation, clinical research, and care consults. The company also released HealthBench Professional, an open benchmark that evaluates LLMs on realistic clinician chat tasks across care consults, documentation, and medical research. OpenAI reports that GPT-5.4 in the Clinicians workspace scored 59.0 on HealthBench Professional versus a human physician baseline of 43.7.

ChatGPT for Clinicians packages product and evaluation choices aimed at clinical workflows and governance. Key product capabilities called out by OpenAI include:

• •a clinical search drawing on millions of peer-reviewed sources and literature indices
• •a deep research mode for structured literature reviews and evidence synthesis
• •reusable templates for referral letters, prior authorizations, and other administrative tasks
• •integrated pathways to earn continuing medical education credit while researching
• •data governance options, including that clinician conversations will not be used to train models, and HIPAA support via a Business Associate Agreement for eligible customers

OpenAI says it developed the product with hundreds of physician advisors and reviewed over 700,000 model responses during testing. The company also published the HealthBench Professional artifact and a technical report describing tasks and scoring. The benchmark measures multi-turn, clinically realistic chat tasks; OpenAI compared GPT-5.4 to models from Anthropic, Google, and xAI in their reported results.

This launch sits at the intersection of productization, regulated-industry deployment, and benchmarking. Making a clinician-focused workspace free for verified U.S. clinicians lowers adoption friction and accelerates real-world usage and feedback loops. The public release of HealthBench Professional attempts to push evaluation toward realistic chat scenarios rather than isolated question-answer tasks, addressing a long-standing mismatch between benchmark conditions and clinical workflows. The claim that GPT-5.4 outperforms physicians on the benchmark is attention-grabbing, but practitioners should interpret that result with caution because OpenAI built both the system under test and the benchmark and controlled the evaluation pipeline. That creates an inherent bias risk; independent replication and third-party evaluations will be necessary to validate performance and safety in deployment.

For ML engineers and data scientists working on healthcare AI, this matters on three fronts: access to clinicians for human-in-the-loop labeling and evaluation will increase as more clinicians use the free workspace; HealthBench Professional provides a more realistic evaluation suite you can adopt or adapt for model comparisons and red-teaming; and enterprise governance patterns, such as conversations-not-used-for-training and BAA-enabled accounts, are emerging defaults you should bake into product design and procurement decisions.

Verify HealthBench Professional results via independent evaluations and check how well GPT-5.4 and other models handle edge cases like hallucinations, rare diseases, and medicolegal wording in documentation. Also watch how hospitals and health systems adopt the free access offer and whether usage uncovers safety or privacy gaps that require product or regulatory changes.