Keep pulling the thread on Karan Singhal.
OpenAI plans to launch ChatGPT Health, a product that connects to electronic medical records and wearables, and a physician-facing "ChatGPT for healthcare" in early 2026.
OpenAI's latest health-focused AI models have achieved performance equivalent to that of an attending physician.
On OpenAI's HealthBench Hard benchmark, model performance has improved from GPT-4.0's initial score of 0% to a current score of approximately 40%.
Approximately 230 million people use ChatGPT for health and wellness-related queries on a weekly basis.
A randomized trial conducted by OpenAI with Kenya's Penda Health System showed a statistically significant improvement in patient outcomes when doctors used an AI co-pilot.
OpenAI plans to make its ChatGPT Health product available globally to all users for free, without advertisements, and without rate limits on its reasoning model.
OpenAI will not use data connected by users within the ChatGPT Health product to train its foundation models.
OpenAI has announced "ChatGPT for Healthcare," a HIPAA-compliant version of its product designed for clinicians that includes features like evidence retrieval from medical guidelines.
OpenAI collaborates with a cohort of approximately 260 physicians to ensure its health AI models produce accurate, robust, and culturally appropriate responses.
OpenAI developed the HealthBench benchmark, which uses 5,000 conversations and 49,000 different criteria to evaluate the performance of large language models in healthcare.
Karin Singhal predicts that 2026 will be the year that using AI becomes a standard part of medical practice, based on positive reception from health systems.
OpenAI has not observed large-scale evidence of its models' chain-of-thought reasoning degrading into "neural ease," a non-human-interpretable internal language.