OpenAI announced on Monday (November 16) an expansion of ChatGPT for Healthcare: computer-use agents will now carry out administrative tasks in electronic health records and payer portals, operating the software on screen — reading, clicking and typing — where no API is available. The announcement also brings improved clinical search and more than 20 new plugins.
What changes in practice
Computer-use agents run multi-step workflows in the systems healthcare teams already use: preparing prior authorizations, closing referrals, planning discharges and coordinating care between visits. In a demo built on synthetic patient data, an agent reviews the chart, drafts an MRI order, an encounter note and a reply to the patient for the clinician to review, then prepares the prior authorization request in the payer portal for a specialist to review.
Results reported by OpenAI
The figures below come from OpenAI itself, not from an independent evaluation — except for HealthAdminBench, described by the company as an independent benchmark from Stanford researchers. On electronic health record inbox tasks, shaped with physician input, reviewers rated 99% of 325 tasks performed by GPT-6 Astra as safe, and physicians rated the work “very helpful” or “extremely helpful” in 93% of 646 assessments. On HealthAdminBench, which measures administrative tasks such as prior authorizations, appeals and medical equipment orders, GPT-6 Astra earned 94.3% of the rubric points and GPT-6 Luna 91.8% — at a cost of $0.085 per task, 98% less than Astra. Both models were evaluated at maximum reasoning effort. In clinical search, physicians reviewed answers to real-world questions across specialties in more than 1,300 reviews and rated the updated experience higher on all eight dimensions measured, including factual correctness, safety and source authoritativeness.
Partners and use cases
OpenAI names Abridge, Ambience, Cigna Healthcare and CVS Health among the organizations developing, piloting or deploying the agents. CVS Health says one agent cleared a months-long document-sending backlog over a weekend, and another handles the system steps for tens of thousands of consults a day; the company is open-sourcing Agent Kit, built with the OpenAI Agents SDK. At Cigna Healthcare, case managers previously had to navigate more than 10 clinical and administrative systems to gather a member’s context — computer use now brings that information together in seconds, according to the company. Abridge is developing agents to follow up on hospital discharges, and Ambience built controls so the agent stops and asks for approval when the chart is ambiguous. Among startups, Assort uses computer use in Codex to close referrals in the Athena system, where the API does not reach.
Plugins and availability
More than 20 new plugins are coming to ChatGPT for Healthcare, including ChartGPT, which brings ChatGPT into the electronic health record workflow. Partners include Aiva, Amion, Assort Health, Availity Extend, Elation Health, Ferry Health, IMO Health, Infinitus Relay, K Health, Luma Health, Metriport, Notable, Oracle HDI, PenguinAI, Photon, Redox, Stedi, Subsalt, TriZetto, Turquoise and UpToDate Expert AI. The features are already available in ChatGPT for Healthcare, ChatGPT for Clinicians and the OpenAI API, subject to workspace settings; administrators must enable the new plugins. ChatGPT for Clinicians is free for eligible, verified clinicians in the United States. For workflows involving protected health information, organizations need eligible services covered by a Business Associate Agreement (BAA) with OpenAI, and organization data is not used to train foundation models, according to the company.
Image credit
Featured image: the OpenAI logo, via Wikimedia Commons (file “OpenAI logo 2025 (wordmark).svg”).
Sources
- Advancing computer-use agents for healthcare teams — OpenAI
- More ways to move care forward with ChatGPT for Healthcare — OpenAI
Transparency: this article was produced with the assistance of artificial intelligence, cross-checked against public posts on X and sources on the internet. Please confirm the information in the original sources listed below.
Por Muse