OpenAI has updated ChatGPT’s health and wellness responses for free users, saying GPT-5.5 Instant now performs near its frontier reasoning models on health-specific evaluations and is better at recognizing urgent situations, asking for missing context, explaining uncertainty, and turning medical information into clearer next steps.
The company announced the health-intelligence update on June 18, describing it as part of a broader push to make ChatGPT more useful for everyday health questions. OpenAI says more than 230 million people use ChatGPT each week for health and wellness topics, including lab results, appointment preparation, insurance questions, habits, and basic attempts to understand what to ask a clinician next.
That scale is what makes the update important. This is not a niche clinical tool being tested in a hospital pilot. It is a change to the default model available to free ChatGPT users, in a category where wrong, overconfident, or poorly contextualized answers can carry much higher stakes than ordinary search or productivity help.
What OpenAI says changed
OpenAI says GPT-5.5 Instant, released in May 2026 and available to free ChatGPT users subject to limits, substantially improved over GPT-5.3 Instant on an aggregate of health evaluations, including HealthBench Professional. Those evaluations look beyond medical trivia and try to measure qualities that matter in real conversations: accuracy, safety, context awareness, completeness, communication, and whether an answer properly escalates someone toward care.
The company also compared model answers with responses written by physicians who had unlimited time and access to the internet but no AI tools. A separate physician panel reviewed 3,500 responses across factors such as accuracy, completeness, instruction following, communication quality, and health decision helpfulness. In that evaluation, GPT-5.5 Instant responses were rated higher than both physician-written responses and older model responses.
OpenAI’s strongest production claim is about factuality. Using privacy-preserving monitors on health-related traffic, the company says the rate of health responses with at least one flagged factuality issue fell 71% over the last two months. That figure is based on production traffic rather than a public clinical trial, so it should be read as OpenAI’s measurement of its own system, not as independent proof that ChatGPT is safe to use as a substitute for medical care.
Why HealthBench matters
The update leans heavily on HealthBench, OpenAI’s health-evaluation suite introduced in 2025. HealthBench was built with 262 physicians who had practiced in 60 countries, spoke 49 languages, and covered 26 medical specialties. The benchmark includes 5,000 realistic health conversations and 48,562 physician-written rubric criteria.
That structure matters because many older medical AI tests looked more like exams than real patient conversations. A useful health assistant has to do more than name a condition. It has to notice when a symptom could be urgent, avoid false certainty when a question is underspecified, tailor advice to local healthcare context, explain tradeoffs plainly, and encourage a user to contact a clinician when the situation calls for it.
OpenAI says more than 260 physicians now support its health-response work across 60 countries, 49 languages, and 26 specialties. They review example responses, identify failure modes, help define ideal behavior, and turn that feedback into rubrics for model evaluation. The company says physicians have reviewed more than 700,000 example responses so far.
The more useful way to read those numbers is not that ChatGPT has become a doctor. It is that OpenAI is treating health answers as a distinct product and safety problem, with separate evaluation, expert review, and production monitoring rather than relying only on general model improvements.
The unresolved safety and privacy question
The health update arrives while OpenAI is under growing scrutiny over user safety and sensitive information. The Associated Press reported that OpenAI received a subpoena from several states as part of a probe into the safety of ChatGPT users, including concerns around minors, people in distress, health data, and personal information. OpenAI told AP it would respond constructively and pointed to safeguards for minors and users experiencing difficult situations.
That regulatory context is not separate from the health rollout. Health conversations are unusually sensitive because users may share symptoms, medications, lab values, insurance details, mental-health concerns, family history, or location-specific care questions. Better answers help, but they also make the product more attractive for exactly the kinds of conversations where privacy, retention, consent, and escalation behavior matter most.
OpenAI’s January launch of ChatGPT Health positioned the product as a dedicated experience for bringing health data and AI guidance together. The June update pushes the general-purpose model closer to that ambition by improving the answers available to free users, not only people using a specialized paid or clinical workflow.
How readers should use the update
For ordinary users, the best use of ChatGPT health answers is preparation, translation, and question-building. It can help explain a lab term, summarize what a discharge instruction might mean, suggest questions to bring to an appointment, compare common treatment options, or help organize symptoms before calling a clinician.
It should not be treated as a diagnosis engine, an emergency triage line, or a replacement for a doctor who can examine a patient, order tests, review a chart, and take legal responsibility for care. If symptoms are severe, sudden, worsening, or potentially urgent, the safer path is still emergency services, an urgent-care line, a clinician, or a local medical advice service.
Users should also be careful with what they paste into any AI system. Before sharing health details, remove information that is not needed for the question, such as full names, exact addresses, account numbers, medical record numbers, or insurance identifiers. When a question involves a child, a mental-health crisis, pregnancy, medication changes, chest pain, breathing trouble, neurological symptoms, or possible poisoning or overdose, ChatGPT should be a supplement at most, not the deciding authority.
A bigger consumer AI test
OpenAI’s health update is a test of whether consumer AI assistants can move into high-stakes personal domains without pretending that benchmark scores settle the trust problem. The company has evidence that GPT-5.5 Instant is giving better health answers than older models, and its physician-led evaluation work is more serious than a generic accuracy claim.
But health is also where the remaining gaps become hardest to tolerate. The real standard is not whether a model can write a polished answer. It is whether it can consistently ask the right follow-up questions, recognize when not to answer confidently, point users toward care at the right time, protect sensitive data, and avoid becoming the only source someone relies on when they need a human professional.
That is why this update is worth watching. GPT-5.5 Instant may make ChatGPT more helpful for millions of health questions each week. It also makes the governance of AI health advice more urgent, because the more useful these systems become, the more likely people are to trust them when the stakes are personal.
