On July 23, 2026, OpenAI opened ChatGPT Health to all U.S. adults across every plan (free, Go, Plus, Pro), on web and iOS. The health mode can now pull data from apps and medical systems straight into a conversation. The announcement came one day after a Florida pastor sued the company, alleging it gave harmful health advice.
Key takeaways
- Available to all U.S. adults across every tier (free, Go, Plus, Pro), on web and iOS
- Data integration: Apple Health, Function Health, MyFitnessPal, plus medical records from Epic and Oracle Health
- 300 million weekly health queries — up from 230 million at the January 2026 test launch
- The smallest model, GPT-5.6-Luna, outperforms GPT-5.5 on the internal HealthBench benchmark
- OpenAI states it does not use user data to train its models
From a separate hub to ordinary chat
Until now the health features lived in a dedicated hub — a separate place you had to enter on purpose. With the change, users can weave connected health data into any conversation, for example asking about an allergen while looking at a recipe. It is a response to real behavior: during testing, 70% of health queries happened outside the dedicated hub.
The integration reaches wide. ChatGPT Health connects to Apple Health, Function Health and MyFitnessPal, and — more importantly — to medical records from Epic and Oracle Health, as well as One Medical. That shifts the assistant from general advice toward analysis of a specific patient's personal data.
Scale: 300 million queries a week
Scale explains why OpenAI treats health as a priority. At the January 2026 test launch, ChatGPT handled roughly 230 million health-related queries a week. Today the figure reaches 300 million — a jump of about a third in half a year. Health has become one of the most common reasons people reach for the chatbot at all.
A smaller model, a better score
The model layer runs on GPT-5.6-Luna, the smallest model in the family. OpenAI says it beats the older GPT-5.5 on HealthBench — the company's internal, open-source benchmark for scoring model answers to health questions. HealthBench is publicly available in the openai/simple-evals repository, alongside tests such as MMLU and MATH. Choosing the smallest model as the default for health signals that OpenAI is optimizing for cost and response speed, not just raw power.
Disclaimers and privacy
OpenAI keeps a clear legal distance. The company notes the service is "not intended for use in the diagnosis or treatment of any health condition" and that users should verify information and consult medical professionals. It also states it does not use user data to train its models and works with physicians on the quality of health answers.
These disclaimers are not cosmetic. The announcement landed one day after a Florida pastor sued OpenAI over harmful health advice. Independent studies have warned for months that chatbots are not a reliable source of medical advice — which has not stopped the largest companies from expanding health features.
Not just OpenAI
The direction is industry-wide. Anthropic launched Claude for Healthcare in January 2026, and Google added an AI-powered health and fitness coach to Fitbit back in August 2025. The difference is distribution: OpenAI puts the health mode into the free plan and a base measured in hundreds of millions of weekly queries, making it the default point of contact for a far broader group than rivals' specialized products.
| Company | Product | Launch |
|---|---|---|
| OpenAI | ChatGPT Health | July 2026 |
| Anthropic | Claude for Healthcare | January 2026 |
| Fitbit health coach | August 2025 |
Why it matters
Putting the health mode into the free plan changes the nature of the risk. As long as the feature was paid and tucked into a separate hub, a narrower, more deliberate group used it. Dropping it into the default conversation of a hundred-million-user base means more people will treat the chatbot's answers as the first, and sometimes only, filter for a health decision.
Integration with Epic and Oracle Health records raises the stakes: the model stops trading in generalities and starts interpreting a specific patient's actual test results. That boosts usefulness, but also the weight of every error.
The legal disclaimers and the no-training-on-user-data pledge serve as a reputational and regulatory shield — especially in the shadow of a fresh lawsuit. The key question is no longer whether people ask AI about health (300 million weekly queries answer that), but who is liable when the answer is wrong.
What's next
- The rollout currently covers web and iOS in the U.S. — no confirmed date for Android or markets outside the U.S.
- The Florida pastor's lawsuit will test AI providers' legal liability for health advice — its course may shape the scope of disclaimers in the product.
- OpenAI says it will keep working with physicians on answer quality — future HealthBench updates will show whether GPT-5.6-Luna keeps its edge.
Sources
- TechCrunch — OpenAI makes ChatGPT Health available to all U.S. users
- GitHub — openai/simple-evals (HealthBench)
