OpenAI says GPT-5.5 Instant, its free default ChatGPT model, now scores higher than physician-written responses on health evaluations, putting a new decision in front of patients, caregivers, clinicians, and healthcare organizations: when should people trust AI health answers, and when do they still need a doctor?

More than 230 million people use ChatGPT each week for health and wellness questions. Many are trying to understand symptoms, lab results, insurance, appointment questions, medical terms, healthier habits, and what to ask next when medical information is confusing or incomplete.

OpenAI says GPT-5.5 Instant is now better at recognizing urgent-care situations, asking for missing context, explaining uncertainty, and making medical information easier to understand. The improvement affects users seeking health explanations, clinicians who may see more AI-informed patients, and healthcare organizations deciding how AI should support patient communication.

In short, OpenAI is saying ChatGPT’s health answers are now more accurate, more complete, and more useful for health decisions than they were before. That does not make ChatGPT a doctor, but it may make ChatGPT more useful for understanding medical information and preparing for conversations with qualified clinicians.

Health AI support means using an AI assistant to explain medical information, organize questions, clarify uncertainty, and help people decide what to ask a healthcare professional next.

A health AI assistant helps users understand medical information, prepare better questions, and recognize when professional care may be needed.

More than 230 million people use ChatGPT each week for health and wellness questions, according to OpenAI. They are asking about lab results, appointment preparation, insurance, healthier habits, and what to ask next when they are trying to make sense of medical information.

At that level of use, the health question has already moved beyond curiosity. People are using ChatGPT to understand symptoms, prepare for conversations with doctors, sort through confusing medical terms, and decide what information they still need before or after seeking care.

Health is different from many other AI use cases because a weak answer can create real risk. A vague response may leave someone confused. An overconfident response may make a problem sound less serious than it is. A better answer has to be accurate, understandable, cautious when the situation is uncertain, and clear about when professional care may be needed.

OpenAI says GPT-5.5 Instant is a major improvement for health questions, shaped by model advances and physician-led evaluation. If more people are using ChatGPT to get health information, what has actually changed, and how is OpenAI measuring better health responses?

OpenAI says GPT-5.5 Instant, the default model available to free ChatGPT users, has made a substantial improvement on health questions. The company says the model is better at recognizing when urgent care may be needed, asking for relevant context, explaining uncertainty, and making complex health information easier to understand, giving more people access to stronger health answers.

On OpenAI’s most challenging health evaluations, GPT-5.5 Instant now performs at a level comparable to its frontier Thinking models. That means OpenAI is saying its default free-access model has reached a similar level of health-answer quality as its more advanced reasoning models on those tests.

OpenAI also tested the model against physician-written responses. Physicians wrote answers to representative health conversations with unlimited time and internet access, but without using AI. A separate panel of physicians then compared those answers with model responses across 3,500 reviewed responses.

The reviewing physicians judged accuracy, communication, completeness, instruction following, and helpfulness for health decisions. According to OpenAI, physicians rated GPT-5.5 Instant as having fewer mistakes than both older models and the physician-written responses.

Those mistakes included failing to account for local healthcare context, missing red flags or referrals to care, and not asking for more information when the answer needed more context. OpenAI’s improvement claim sounds strong, and the comparison depends on how the company defines a “better” health response.

OpenAI measures health progress through health-specific evaluations, including HealthBench and HealthBench Professional. Those evaluations use realistic health conversations and physician-written rubrics to judge whether a response is accurate, safe, complete, understandable, aware of context, and clear about when medical care may be needed.

HealthBench evaluates 5,000 realistic, multi-turn health conversations between a model and an individual user or healthcare professional, with responses scored against conversation-specific rubrics created by 262 physicians. HealthBench Professional focuses more narrowly on real clinician chats with ChatGPT for Clinicians across 3 common clinical use cases: care consults, writing and documentation, and medical research.

Health-answer quality cannot be measured only by whether a response contains the right facts. A useful answer may need to ask for missing information before giving guidance, explain uncertainty without overstating confidence, simplify medical language, and tell someone when a symptom needs professional attention.

OpenAI says its health evaluations are shaped by a global network of more than 260 physicians across 60 countries, 49 languages, and 26 medical specialties. Those physicians help define what a good health response should look like across everyday wellness questions, complex clinical situations, and real-world patient or clinician use.

Physicians have reviewed more than 700,000 example model responses so far, according to OpenAI. Their feedback informs the rubrics and evaluation criteria OpenAI uses to judge whether responses are accurate, safe, clear, complete, appropriately cautious, and useful. They also help identify where an answer may miss important context, sound too confident, need clearer next steps, or more directly encourage someone to seek medical care.

OpenAI’s benchmark charts show how the company is applying that standard. In one evaluation, GPT-5.5 Instant received higher scores than physician-written responses and GPT-4o across 5 criteria: accuracy, clear communication, completeness, following instructions, and health decision helpfulness.

The largest gaps appeared in completeness and health decision helpfulness:

Because OpenAI built the evaluation system it is reporting, the physician role behind that system becomes more important. The benchmark results show how OpenAI is measuring better health responses, but they are only one part of the evidence the company is using to argue that ChatGPT’s health answers are improving.

OpenAI’s cost-performance chart makes the access point clearer. GPT-5.5 Instant reaches health evaluation performance close to OpenAI’s 5.5 Thinking models, while costing far less to run per 1,000 examples. If OpenAI can deliver stronger health answers through a cheaper, faster default model, those improvements can reach more users instead of staying limited to higher-cost reasoning models.

OpenAI also shows what that improvement looks like in an example health question: “Why might a doctor recommend an MRI before a steroid injection for sciatica?” The older model response explained that an MRI can help identify the cause of sciatica, rule out other conditions, and guide the injection.

The GPT-5.5 response gave a more complete answer. It explained that the injection may be safer and more useful when a doctor knows exactly what is compressing or irritating the nerve and where. It then walked through several possible reasons for the MRI, including confirming the cause of sciatica, choosing the correct injection level and side, selecting a safer injection approach, checking for red flags, deciding whether surgery or another treatment should be considered, and avoiding an injection that is unlikely to help.

The GPT-5.5 answer also added 3 things that are easy to miss in a shorter response. It said MRI is not always required for every case of sciatica. It cited outside medical sources for readers who wanted to learn more. It gave the user a specific question to ask the doctor: “What are you looking for on the MRI, and how would the result change the injection plan?”

Example: Why might a doctor recommend an MRI before a steroid injection for sciatica?

GPT-4 model response:
Before a steroid injection for sciatica, a doctor might recommend an MRI (Magnetic Resonance Imaging) to get a detailed image of the patient's spine and the surrounding tissues. This helps in identifying the exact location and cause of the sciatica, such as a herniated disk or spinal stenosis.

An MRI can also rule out other conditions that could be causing the pain. Additionally, it helps in guiding the physician during the injection procedure, ensuring that the medication is accurately delivered to the area causing the pain.

GPT-5.5 Instant response:
A doctor may recommend an MRI before a steroid injection for sciatica because the injection is more likely to be useful—and safer—when they know exactly what is compressing or irritating the nerve and where.

Common reasons include:

That said, MRI is not always required for every case of sciatica. In people without red flags, many cases improve with time, physical therapy, and medication. Some studies have questioned whether routine MRI before lumbar epidural steroid injection always improves outcomes, but guidelines and many clinicians still favor imaging when symptoms persist and an injection or surgery is being considered. (pmc.ncbi.nlm.nih.gov)

You should ask your doctor: “What are you looking for on the MRI, and how would the result change the injection plan?” That usually clarifies whether the MRI is for diagnosis, safety, targeting, or deciding whether an injection is appropriate at all.

Cited:

The GPT-5.5 response gave a more medically useful answer. It explains the medical reasoning, shows where uncertainty remains, cites outside medical sources, and gives the user a concrete question to bring back to the doctor.

OpenAI uses another measure beyond benchmark testing: production traffic. The company says it uses privacy-preserving monitors to track possible factuality issues in health responses across real ChatGPT usage.

Based on a comparison of recent health traffic across billions of messages per week, OpenAI says the rate of responses with at least 1 flagged factuality issue has fallen by 71% in the last 2 months. That gives OpenAI a real-world signal beyond controlled benchmark scores.

That 71% figure measures one specific problem: whether a health response was flagged for a possible factuality issue. It does not measure whether ChatGPT improved patient outcomes, made correct diagnoses in clinical care, recommended safe treatments, or performed like a licensed medical professional.

Fewer flagged factuality issues can make ChatGPT more useful for explaining medical information, clarifying confusing terms, preparing questions for a doctor, and helping users recognize when professional care may be needed.

OpenAI is improving health responses because people are already asking ChatGPT medical questions at massive scale. Better answers can help users understand health information, but ChatGPT is not a doctor. It should not be used to diagnose a condition, choose a treatment, delay urgent care, or replace advice from a qualified clinician.

OpenAI’s evidence gives users more reason to trust ChatGPT than before for health explanation, preparation, and follow-up questions. GPT-5.5 Instant is scoring higher on physician-reviewed health evaluations, showing fewer flagged factuality issues in real-world health traffic, and giving more complete answers in examples such as the MRI-before-injection question.

GPT-5.5 Instant may be getting better at the parts of health support where an AI assistant can help: explaining confusing terms, organizing information, helping a user understand why a doctor might recommend a test, asking for missing context, showing uncertainty, and helping someone prepare for a conversation with a clinician. It should not decide whether symptoms are safe to ignore, whether a diagnosis is correct, or whether a treatment is right.

ChatGPT is not a doctor and should not be used as one. But stronger health answers can still help people who need clearer information outside a doctor’s office, including someone trying to understand a symptom late at night or decide whether to call a clinician.

OpenAI has not shown that ChatGPT can replace clinical judgment, and it is not reporting independent validation of ChatGPT as medical care.

So the answer is yes: users may have more reason to trust ChatGPT’s health answers than they did before, especially when they need help understanding medical information, preparing questions, clarifying options, or recognizing when professional care may be needed. They should not trust it to diagnose a condition, choose a treatment, or replace a doctor’s judgment.

Q: Can I trust ChatGPT more for health questions now?
A: OpenAI’s evidence gives users more reason to trust ChatGPT for health explanation, preparation, and follow-up questions than they had before. GPT-5.5 Instant scored higher on physician-reviewed health evaluations, showed fewer flagged factuality issues in real-world health traffic, and gave more complete answers in OpenAI’s examples. It should still not be used to diagnose a condition, choose a treatment, delay urgent care, or replace a qualified clinician.

Q: What did OpenAI say about GPT-5.5 Instant and health answers?
A: OpenAI says GPT-5.5 Instant, the free default ChatGPT model, now scores higher than physician-written responses on its health evaluations. The company says the model is better at recognizing urgent-care situations, asking for relevant context, explaining uncertainty, and making complex health information easier to understand.

Q: How many people use ChatGPT for health questions?
A: OpenAI says more than 230 million people use ChatGPT each week for health and wellness questions. Users ask about lab results, appointment preparation, insurance, healthier habits, symptoms, medical terms, and what to ask next.

Q: Is ChatGPT better than a doctor?
A: No. OpenAI says GPT-5.5 Instant scored higher than physician-written responses in its health evaluations, but that does not mean ChatGPT can replace a doctor. The comparison measured written answers to representative health conversations. It did not show that ChatGPT can diagnose patients, choose treatments, or provide clinical care.

Q: How does OpenAI test whether ChatGPT gives better health answers?
A: OpenAI uses health-specific evaluations called HealthBench and HealthBench Professional. HealthBench evaluates 5,000 realistic, multi-turn health conversations using physician-written rubrics. The rubrics judge whether an answer is accurate, safe, complete, understandable, aware of context, and clear about when medical care may be needed.

Q: What makes GPT-5.5 Instant better for health questions?
A: OpenAI says GPT-5.5 Instant gives more complete answers, asks for missing context, explains uncertainty, identifies possible red flags, and makes medical information easier to understand. In the sciatica MRI example, GPT-5.5 Instant explained why an MRI could affect safety, injection targeting, treatment choice, red-flag screening, and whether an injection was likely to help.

Q: What does OpenAI’s 71% drop in health factuality issues mean?
A: OpenAI says the rate of health responses with at least 1 flagged factuality issue fell by 71% in recent real-world ChatGPT traffic over the last 2 months. That suggests fewer possible factuality problems in health answers. It does not prove better patient outcomes, correct diagnoses, safe treatment choices, or medical reliability in every situation.

OpenAI’s health update shows ChatGPT becoming a more serious health information tool for people trying to understand medical questions outside the doctor’s office. Many users turn to AI because they leave appointments, portals, lab reports, insurance forms, or online searches with questions they still cannot answer clearly.

ChatGPT is becoming more useful as a health literacy and preparation tool. OpenAI’s evidence supports improvement in accuracy, completeness, communication, and health decision helpfulness, especially when the user needs a clearer explanation or a better question to ask a clinician. It does not support treating ChatGPT as medical care.

Patients and caregivers should care because many health decisions begin with confusion: what a term means, why a doctor ordered a test, whether a symptom sounds concerning, or what question to ask next. Clinicians should care because patients may arrive with more AI-assisted questions. Healthcare organizations should care because if patients need ChatGPT to understand basic medical communication, the healthcare system may not be explaining enough in the first place.

ChatGPT is already being used for health questions at massive scale, with OpenAI reporting more than 230 million weekly health and wellness users and health traffic across billions of messages per week. Better answers can reduce confusion, but high-stakes health questions still carry risk when an AI answer sounds confident, misses context, or is used instead of professional care.

Readers have to decide where ChatGPT belongs in their health decisions. It can help explain a doctor’s recommendation, prepare appointment questions, clarify medical language, and organize follow-up questions. It should not decide whether a symptom is urgent, whether a diagnosis is correct, or whether a treatment is right.

In short, OpenAI’s evidence gives users more reason to trust ChatGPT for understanding health information than they had before. It does not give users a reason to trust ChatGPT as a doctor. ChatGPT is most useful when it helps people ask better questions, not when it becomes the final answer.

Health AI earns trust when it helps people understand their care clearly enough to ask better questions, recognize warning signs, and know when a doctor needs to be part of the answer.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing support, AEO/GEO/SEO optimization, image concept development, and editorial structuring support from ChatGPT, an AI assistant. All final editorial decisions, perspectives, and publishing choices were made by Alicia Shapiro.