OpenAI Launches ChatGPT Health, Claims Models Reason ‘Better Than Clinician Level’

OpenAI is rolling out ChatGPT Health to all US users on Thursday, allowing anyone aged 18 and older to connect their medical records and health-tracking information to the chatbot. The feature, which will appear in the main chat experience on web and iOS, aims to offer personalized health insights based on user-provided data from providers, Apple Health, Weight Watchers, MyFitnessPal, and Function.

During a briefing, OpenAI’s vice president of health product, Ashley Alexander, made a striking claim: the company’s models “are now capable of reasoning at levels that are better than clinician level.” However, OpenAI health lead Karan Singhal quickly tempered that statement, saying he would “temper” the claim and pointing to individual studies — such as one from Harvard and Stanford — that “have been pointing in that direction.” The conflicting messaging raises questions about how ready these models really are for clinical decision support.

The launch arrives against a backdrop of real-world risk: on Wednesday, a Florida pastor sued OpenAI, alleging that ChatGPT provided “extremely dangerous medical recommendations” that led him to delay treatment for a pulmonary embolism. OpenAI is careful to note in its materials that ChatGPT Health “supports, not replaces, professional care,” and that all health data receives additional encryption protection.

ChatGPT Health runs on OpenAI’s latest GPT-5.6 Sol model, which the company touts as its “strongest model yet for health.” Users will be asked for permission before the chatbot accesses their health data, and they can review or change what types of information are connected — including post-visit notes, lab results, and medications.

What Developers Should Know

For developers building health-related apps or integrations, this launch signals that OpenAI is serious about handling sensitive health data with extra security layers. The API access to ChatGPT Health is not yet detailed, but the underlying model capabilities — including the controversial “better than clinician” reasoning — could eventually be exposed through the OpenAI API. Developers should watch for updates on HIPAA compliance and data handling policies before integrating health-related features.

The incident with the Florida pastor also underscores the need for robust safety guardrails when using LLMs in health contexts. OpenAI’s encryption promises and permission-based access are steps in the right direction, but the lawsuit shows that even with disclaimers, users may rely on model outputs in harmful ways.