OpenAI has concluded a two-year evaluation project for its ChatGPT Health interface. The initiative involved over 260 practicing physicians from 60 countries, representing dozens of medical specialties. These experts reviewed more than 600,000 model responses across 30 distinct medical fields to refine the system's performance.
The product was officially announced on January 7, 2026. By July, it became available to selected users over 18 in the United States via web and iOS platforms. The interface allows users to integrate personal medical records and fitness application data to contextualize the information provided by the AI.
The feedback from medical professionals directly influenced how the system generates recommendations and handles potentially hazardous or ambiguous queries. OpenAI explicitly states that the tool is not intended for diagnosis and does not replace professional medical care. Furthermore, data from these health-related dialogues are not used to train the company's base models.
In conjunction with this project, the company developed HealthBench, a benchmark designed to assess the quality of medical responses. Many of the physicians involved in the initial evaluation also contributed to the creation of this tool. OpenAI emphasizes that this expert assessment is distinct from clinical validation, and no data regarding patient outcomes has been released.
