OpenAI Opens ChatGPT Health to All U.S. Adults, But Free Users Get a Weaker Model

OpenAI switched on ChatGPT Health for every U.S. user aged 18 and over on July 23, 2026, moving a feature that had been limited to a small test group in January into general availability across Free, Go, Plus, and Pro tiers on web and iOS.

The expansion lets people link medical records from hospital systems like Epic and Oracle Health, alongside consumer data from Apple Health, Function Health, MyFitnessPal, Weight Watchers, and One Medical. Once connected, ChatGPT can pull medications, lab results, recent visits, sleep patterns, and activity logs directly into a conversation, provided the user has granted permission.

That data now flows into any chat, not just a dedicated section. OpenAI redesigned the experience after discovering that 70% of health-related questions during testing happened outside the standalone health hub, meaning most people were already asking about symptoms or medications in the regular chat window. The Health tab now functions more as a control panel: a place to manage connected accounts, review synced data, and revisit past health conversations.

The rollout comes with an unadvertised split in model quality. Paying subscribers get responses from GPT-5.6 Sol, which OpenAI calls its strongest model for health tasks, while free-tier users are routed to GPT-5.5 Instant. On HealthBench Professional, OpenAI's own benchmark for clinical usefulness, the gap is stark.

Health decision helpfulness follows the same pattern, at 83.0% for the paid model against 50.8% for the free one. TechCrunch's reporting refers to the premium model as "GPT-5.6-Luna," while OpenAI's own blog, The Verge, and The Decoder all use "GPT-5.6 Sol," a naming discrepancy that appears to be a reporting error rather than two distinct models.

OpenAI says weekly health-related queries on ChatGPT have climbed to over 300 million, up from 230 million in January, underscoring how much medical-adjacent usage the company is now trying to formalize rather than leave unmanaged. The company says all conversations are encrypted in transit and storage, with an added encryption layer for health data, and that none of it is used for model training.

Legal exposure arrived a day early. On July 22, a Florida pastor sued OpenAI, alleging that ChatGPT gave him "extremely dangerous" medical guidance that led him to delay treatment for a pulmonary embolism.

OpenAI maintains in its terms that its services are not intended for use in the diagnosis or treatment of any health condition.

That disclaimer sits awkwardly next to a product now designed to ingest lab results and medication histories at scale, and it will likely shape how the lawsuit, and any that follow, gets argued.