OpenAI is adding computer-use agents, improved clinical search and more than 20 new healthcare plugins to ChatGPT for Healthcare, it said in a product post on 16 November. In a companion research post, it says that on HealthAdminBench, an independent benchmark from Stanford researchers for administrative work such as prior authorizations, its cheaper GPT-6 Luna model earned 91.8% of available rubric points at 8.5 cents per task, against 94.3% for GPT-6 Astra, while costing 98% less.
Our number: Luna kept about 97% of Astra’s score for about 2% of its cost per task, by our arithmetic (91.8 divided by 94.3, and 100% minus 98%). OpenAI names Abridge, Ambience, Cigna Healthcare and CVS Health as organisations developing, piloting or deploying the agents, and says the capabilities are available today in ChatGPT for Healthcare, ChatGPT for Clinicians and the OpenAI API.
Business Pill · AI Agents
An evergreen Business Engineer explainer on how AI agents move from answering to doing. It teaches the general idea only and does not discuss OpenAI’s healthcare launch.
The key insight: As we read OpenAI’s posts, the price of an agent task is becoming the selling point for healthcare paperwork: on HealthAdminBench GPT-6 Luna scored 91.8% at 8.5 cents a task, about 97% of GPT-6 Astra’s 94.3% for about 2% of its cost, by our arithmetic, while more than 20 new plugins bring the tools into the same conversation.
What OpenAI Launched
The product post groups the update in three parts. Computer-use agents carry out multi-step work across existing healthcare software by reading screens, clicking, typing and moving between applications. Trusted clinical search, which finds, evaluates and cites medical evidence, has been improved. And OpenAI is “expanding our healthcare plugin ecosystem with more than 20 new healthcare plugins from leading companies across clinical care, care coordination, and business operations”.
The post’s worked examples name 11 plugin companies, by our count: Luma Health, TriZetto, Metriport, UpToDate, AIVA, Amion, IMO Health, Stedi, Turquoise Health, Assort Health and Ferry Health. They cover tasks such as checking insurance eligibility, finding a discharge summary, pulling up a hospital policy and reviewing referral volume.
ChatGPT for Clinicians is free for eligible, verified clinicians in the U.S., OpenAI says, and both products offer a Business Associate Agreement to support HIPAA-compliant workflows. Healthcare customers get the agents and the improved search today, subject to workspace settings, and an administrator has to enable the new plugins. Organisations can buy ChatGPT for Healthcare directly or through an AWS Marketplace private offer, via OpenAI’s sales team.

The Numbers in OpenAI’s Evaluations
OpenAI tested GPT-6 Astra on tasks clinicians handle in their EHR inboxes, shaped by physician input. It says “99% of 325 reviewed tasks were rated safe”, and that physicians picked “very helpful” or “extremely helpful” in 93% of 646 assessments, on a five-point scale.
On HealthAdminBench, both models were run at maximum reasoning effort, the research post says. OpenAI’s reading is that the results “show progress toward capable, affordable agents for high-volume administrative workflows”. For clinical search, physicians reviewed responses to real-world clinical questions in a blind evaluation; in over 1,300 reviews they rated the updated experience more highly across all eight dimensions evaluated, including factual correctness, safety, actionability and source authoritativeness, OpenAI says.

Who Is Putting the Agents to Work
CVS Health gives production results in the research post. In one production workflow, an agent cleared a months-long document-sending backlog over a weekend, and another handles the system steps for tens of thousands of consults each day, OpenAI says. CVS Health is open sourcing its Agent Kit, built with OpenAI’s Agents SDK.
Cigna Healthcare case managers previously had to navigate more than 10 separate clinical and administrative systems to gather context on a member, the post says; computer use can now help bring that information together in seconds. Ambience’s agent prepares a prior-authorization request and, when the record is ambiguous, stops and asks the clinician before submitting. Abridge is developing agents for a planned inpatient use case, including discharge barriers.
How It Compares
Our cross-read: six days earlier Google Cloud launched Gemini for healthcare in U.S. preview, and its launch post named 27 connectors by our count, against OpenAI’s “more than 20” new plugins. Both companies pitch agents that work inside the systems teams already use.
Menlo Ventures, whose report says “Administrative AI still commands the biggest budgets”, estimates healthcare AI spend at $3.7 billion in 2026, with clinical copilots at $550 million, up 5.5x, and coding and billing at $600 million. Start-ups are working the same phone-and-portal problem: health AI start-up August raised $12 million for an agent that calls doctors’ offices for patients.
The Structural Read
As we read the launch, the plugins and the computer-use agents split the work: plugins reach systems that have an integration, and the agents operate screens where, in Ambience’s words in the post, APIs don’t reach.
As we read the cross-read with our earlier reporting, OpenAI and Google are pitching the same shape of product to health systems within a week: Google’s Gemini for healthcare named 27 connectors by our count, and OpenAI now says more than 20 new plugins.
OpenAI, Advancing computer-use agents for healthcare teams, 16 November 2026
“These results show progress toward capable, affordable agents for high-volume administrative workflows.”
Three Implications
FOR HEALTH SYSTEMS OpenAI says organisations can configure which applications agents can access, add review and approval steps, and record audit logs, and must use services covered by their Business Associate Agreement for protected health information.
FOR PAYERS AND PROVIDERS OpenAI names authorization requests, referrals, discharge planning and outreach between visits as the workflows; separately, the post says Cigna Healthcare case managers previously had to navigate more than 10 separate systems to gather context on a member before outreach.
FOR HEALTH AI START-UPS OpenAI says it is inviting more healthcare companies to build plugins, and CVS Health is open sourcing its Agent Kit built with OpenAI’s Agents SDK.
The Business Engineer Lens
This story maps onto the Business Engineer framework Enterprise AI: From Software to Substrate.
The framework says: “The value proposition shifts from features to integration depth.”
As we read OpenAI’s launch, the plugins and the computer-use agents are both bids for integration depth inside health systems’ existing software; the framework predates the launch and does not discuss it.
What Is Not Established
The evaluation figures are OpenAI’s own. The posts we read give Luna’s cost per task and the 98% gap, not a per-task cost for Astra, so the 2% is derived from OpenAI’s rounded percentage. HealthAdminBench scores here are as OpenAI reports them; we did not rerun them.
The posts do not list all of the more than 20 plugins, and they give no customer counts, usage figures or list price for ChatGPT for Healthcare in the text we read. The CVS Health and Cigna results are described by OpenAI, without the underlying data.
The Bottom Line
OpenAI is putting computer-use agents, better clinical search and more than 20 new plugins into ChatGPT for Healthcare. On a Stanford administrative benchmark, GPT-6 Luna scored 91.8% at 8.5 cents a task, about 97% of Astra’s 94.3% for about 2% of the cost, by our arithmetic. CVS Health, Cigna, Ambience and Abridge are the named organisations working with it.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. The launch details and figures come from two OpenAI posts published on 16 November 2026, “More ways to move care forward with ChatGPT for Healthcare” and “Advancing computer-use agents for healthcare teams”, which we read in full, and from OpenAI’s post on X announcing them. The HealthAdminBench scores and the physician evaluations are as OpenAI reports them. The comparisons with Google, Menlo Ventures and August come from our own earlier reporting.
The 97%, the about 2% and the count of named plugin companies are our arithmetic. We did not contact OpenAI for this piece.
Sources: OpenAI: More ways to move care forward with ChatGPT for Healthcare (16 Nov 2026) · OpenAI: Advancing computer-use agents for healthcare teams (16 Nov 2026)









