This week, two OpenAI executives sat through the same briefing and offered conflicting accounts of the same product. One claimed the company’s models can now out-reason a physician. When the other was asked to substantiate that, they began qualifying it.
Ashley Alexander, OpenAI’s vice president of health product, stated that the company’s models “are now capable of reasoning at levels that are better than clinician level.” It’s a bold assertion — precisely the sort of thing that warrants a second opinion.
That second opinion came from within the company itself.
Pressed by reporters for specifics on how the models measure up to human clinicians, OpenAI health lead Karan Singhal said he would “temper” the statement. His more restrained version: “there have been individual studies that have been pointing in that direction,” and he cited one out of Harvard and Stanford. But a study that points in a direction is hardly the same as outperforming your doctor — and Singhal appeared to recognize as much.
A lawsuit arrives the day before launch
The timing could hardly be worse. On Wednesday, a Florida pastor sued OpenAI, alleging that ChatGPT handed him “extremely dangerous medical recommendations” that caused him to delay treatment for a pulmonary embolism. The very next day, the company is expanding access to the exact feature designed around medical guidance.
OpenAI’s own messaging attempts to walk that tightrope. It maintains that ChatGPT Health “supports, not replaces, professional care.” Set that beside “better than clinician level,” and the company appears to be making both arguments simultaneously.
What actually ships Thursday
ChatGPT Health rolls out to everyone in the US on Thursday, allowing more users to link medical records and health-tracking data to the chatbot. The most immediate change is more modest than the headline suggests: you can now query your connected health information directly within the main chat rather than switching over to a separate Health tab.
ChatGPT requests permission before drawing on that data to tailor a response. According to the company, “all conversations are encrypted at rest and at transit,” and whatever you link through Health “receives additional encryption protection.”
You decide what gets shared. That includes post-visit notes, notes from your care team, lab results and medications, sourced from health providers or from services such as Apple Health, Weight Watchers, MyFitnessPal and Function. All of it can be reviewed and modified afterward.
The model beneath it
Everything here runs on OpenAI's newest models. The launch follows by weeks the debut of GPT-5.6 Sol, which the company bills as its “strongest model yet for health.” ChatGPT Health taps those latest models under the hood.
It’s open to logged-in users aged 18 and over on web and iOS, spanning the free, Go, Plus and Pro plans.
Link your labs if you’d like. But when a vice president says the model out-reasons a clinician while the health lead declines to endorse that in the same room, the meaningful takeaway isn’t the marketing pitch. It’s which of those two people you’d trust with your pulmonary embolism.