OpenAI rolls out Health in ChatGPT with better model for paid users

OpenAI is rolling out Health in ChatGPT to U.S. users aged 18 and older, expanding a feature it introduced and tested in January. The rollout puts personal health data inside ChatGPT for U.S. adults and creates a quality split by tier: free users get GPT-5.5 Instant, while paying subscribers get GPT-5.6 Sol, which OpenAI says performs better on its health benchmark.

The company said users can connect Apple Health, medical records and wellness apps to ChatGPT. The feature is meant to help people review lab results, prepare for medical appointments and analyze activity or sleep data. OpenAI said connected health data will not be used for model training or advertising. It did not disclose a new price for access to the higher-end model through the health feature.

What is Health in ChatGPT?

Health in ChatGPT is OpenAI’s attempt to make the chatbot a more structured place for health-related questions and personal health data, rather than leaving users to paste symptoms, lab results or device metrics into a general chat. The company says the feature can be used inside any ChatGPT conversation, while a separate Health section remains available for managing connected data and reviewing past health chats.

That design changed after OpenAI’s early testing. The company said more than 70% of participants asked health questions outside the dedicated Health section because switching into it was inconvenient. Making health tools available in normal conversations should reduce friction, but it also raises the stakes for model behavior because health queries are likely to appear in broader, less controlled chats.

OpenAI says more than 300 million people ask ChatGPT health questions each week, up from 230 million in January. The company also said more than 260 physicians helped develop the health features. Those figures explain why OpenAI is formalizing the use case: health is already one of ChatGPT’s largest consumer categories, whether or not the product is positioned as medical software.

Why do paid users get a different health model?

OpenAI is using GPT-5.5 Instant for free users and reserving GPT-5.6 Sol for subscribers. According to OpenAI’s HealthBench Professional results, GPT-5.6 Sol outperformed GPT-5.5 Instant, GPT-4o and physician-written answers across the benchmark’s categories. The widest gaps cited by OpenAI were in completeness, where GPT-5.6 Sol scored 88.0 versus 53.2%, and health decision helpfulness, where it scored 83.0 versus 50.8%.

Benchmark scores do not settle the clinical question. Tests measure answers in controlled settings, while doctors work with patient history, physical exams, nonverbal cues, colleagues and tools that are not captured by a chatbot prompt. OpenAI also says ChatGPT can make mistakes and cannot replace medical advice.

The risk is not abstract. In the RadLE 2.0 radiology benchmark, none of the 16 AI models tested matched human radiologists. The reported weakness was overconfident error: chatbots gave incorrect findings with high confidence instead of recognizing uncertainty, while radiologists were better at saying when they did not know.

Other research has shown more useful results. MIRA, a system for electronic health records, and AMIE both performed about as well as primary care doctors in simulated consultations, according to studies cited in reports on the systems. That points to the more plausible near-term role for health AI: support for routine analysis and preparation, with clinicians retaining responsibility for diagnosis and treatment.

OpenAI has not said when, or whether, Health in ChatGPT will launch in Europe. When the company announced the feature in January, it excluded the European Economic Area, Switzerland and the United Kingdom. Data privacy requirements and possible treatment under the EU AI Act remain obvious hurdles for a product that connects consumer AI to health records.

This story draws on original reporting from The Decoder.