ToolsNews
OpenAI brings computer-use agents to healthcare teams
OpenAI opened computer-use agents to healthcare teams on Monday through the API, ChatGPT for Healthcare and ChatGPT for Clinicians. On Stanford's HealthAdminBench, GPT-6 Luna scored 91.8 percent at 8.5 cents per task.
· 4 min read

What was announced?
OpenAI introduced its healthcare computer-use package in two posts on Monday (OpenAI, announcement). A computer-use agent clicks, types and switches between applications, so it can complete multi-step workflows in systems that have no API. That is common in healthcare: the data sits in the medical record, but someone still has to copy it into an insurer's portal by hand.
The capability is available today in the OpenAI API, in ChatGPT Work and Codex, for ChatGPT for Healthcare customers, and in ChatGPT for Clinicians, which is free for verified US clinicians. Access controls, approval steps and audit logs are part of the package.
What does a workflow look like?
The demo uses synthetic patient data. A patient messages their clinician about lower back pain, and the agent reviews visit notes, physical therapy records and prior imaging in the EHR. It drafts an MRI order, an encounter note and a reply to the patient for the clinician to approve. From the same records it then prepares a prior authorization request in the insurer's portal for an authorization specialist to review before submission.
What do the numbers show?
OpenAI evaluated GPT-6 Astra on clinical EHR inbox tasks shaped by physician input. Reviewers rated 99 percent of 325 tasks clinically safe, and physicians chose "very helpful" or "extremely helpful" in 93 percent of 646 assessments.
For administrative work, OpenAI used HealthAdminBench, an independent benchmark from Stanford researchers covering prior authorizations, appeals and medical equipment orders. GPT-6 Astra earned 94.3 percent of rubric points and GPT-6 Luna 91.8 percent, while Luna cost 98 percent less at 8.5 cents per task. Both ran at maximum reasoning effort. Note that OpenAI reported these numbers, and the clinical evaluation is its own.
Who is using it?
CVS Health automates repetitive work across clinical systems: one agent cleared a months-long document-sending backlog over a weekend, another handles system steps for tens of thousands of specialist consults a day. It verifies actions with screenshots and action logs under human oversight, and it is open sourcing its Agent Kit, built on the OpenAI Agents SDK.
Cigna Healthcare case managers used to search more than ten separate systems to assemble a member's context, and computer use now gathers it in seconds. Cigna says it went from concept to deployment in under eight weeks. Ambience's agent stops on ambiguous records, asks the clinician and waits for approval before submitting, while health systems get a step-by-step audit log. Abridge is working on agents that resolve discharge barriers in parallel.
Clinical search and 20 new plugins
ChatGPT for Healthcare also gets improved clinical search that finds, evaluates and cites medical evidence better. In a blind evaluation with more than 1,300 physician reviews, the updated version scored higher on all eight dimensions, from factual correctness and safety to source authoritativeness.
More than 20 new healthcare plugins arrive as well, including UpToDate Expert AI, Metriport, Stedi, Redox, Oracle HDI and TriZetto. They let teams check eligibility, pull hospital protocols or fetch discharge summaries inside one conversation.
What it means in practice
For developers, computer use is no longer limited to code review or testing (see Greptile's approach). There is now a reference case in a heavily regulated field. The HealthAdminBench results also suggest the expensive model is not always needed: for high-volume admin work, Luna offers far better value. That fits OpenAI's push toward cheap computer use we saw with GPT-6.1 Terra.
One condition is strict: protected health information may only flow through services covered by a Business Associate Agreement, with the required security and retention settings. OpenAI says organization data is not used to train foundation models. For European teams, HIPAA alone is not enough; GDPR and the AI Act's healthcare rules apply too.
Produced with the help of AI, reviewed by an editor. How we make our articles
Reviewed and edited by the Neural Daily desk. About


