Improved Health Intelligence in ChatGPT

OpenAI has significantly strengthened ChatGPT's medical response capabilities through GPT-5.5 Instant. With more than 230 million users asking health-related questions every week, the company focused on improving the model's accuracy and safety.

The new model has improved abilities to recognize situations requiring urgent care, clearly explain uncertain information, and convey complex medical information in an easy-to-understand way. In particular, it showed performance comparable to the existing Thinking model on major medical evaluations including HealthBench Professional.

Key improvements are as follows:

• Physician-led evaluation: A global network of physicians reviews the model's answers, defining ideal behavior patterns and error types to measure performance.
• Improved accuracy and safety: Analysis of real-world service data showed that the rate of responses with factuality issues decreased by 71% over the past 2 months.
• Clinical usefulness: Compared to answers written by physicians, GPT-5.5 Instant recorded better performance than both the existing model and physicians' answers in terms of reflecting local medical context, detecting red flags, and requesting additional information.