Improving health intelligence in ChatGPT

ChatGPT's health intelligence has received a significant upgrade with the introduction of GPT-5.5 Instant. This advancement brings frontier health capabilities to a wider audience, benefiting over 230 million people who use ChatGPT weekly for various health and wellness inquiries, from understanding lab results to preparing for doctor's appointments.

GPT-5.5 Instant demonstrates substantial improvements in several key areas. It is now more adept at recognizing situations requiring urgent care, asking for relevant contextual information, accurately explaining uncertainty, and simplifying complex medical data. On challenging health evaluations, GPT-5.5 Instant performs comparably to OpenAI's most advanced "Thinking" models.

These enhancements are a result of both model advancements and rigorous physician-led evaluations. A global network of medical professionals plays a crucial role in defining optimal responses in real-world health scenarios. They review model outputs, describe ideal behaviors, and identify potential failure modes, ensuring continuous improvement in ChatGPT's health-related responses.

Measuring this progress involves health-specific evaluations such as HealthBench and HealthBench Professional, which utilize realistic health conversations and physician-written rubrics. These assessments gauge accuracy, safety, communication effectiveness, context awareness, completeness, and appropriate escalation. GPT-5.5 Instant shows performance akin to the latest frontier models across these evaluations, marking a substantial improvement over its predecessor, GPT-5.3 Instant.

Further validation comes from physician comparisons. In a study comparing GPT-5.5 Instant responses with those written by physicians (without AI assistance) and older models, GPT-5.5 Instant consistently received higher ratings. Physicians noted fewer failure modes in GPT-5.5 Instant's responses, such as better tailoring to local healthcare contexts, improved identification of red flags, and more effective solicitation of additional user context when necessary.

The real-world impact of these improvements is evident in production traffic data. Privacy-preserving monitors show a 71% reduction in the rate of responses with at least one flagged factuality issue over the last two months, based on billions of messages weekly. This demonstrates ChatGPT's enhanced ability to provide accurate and helpful health guidance, better manage uncertainty, and recognize when urgent attention is needed.