OpenAI’s GPT-5.5 Instant Now Default in ChatGPT as 230 Million Weekly Health Questions Drive AI Upgrade

Every week, more than 230 million people turn to ChatGPT with health questions. That’s a staggering number — and it’s the backdrop to OpenAI’s decision to push its new GPT-5.5 Instant model out as the default across ChatGPT, including to users on the free tier, effective immediately.

The company announced the update in a post on its health intelligence blog, framing GPT-5.5 Instant as a step forward not just in general capability but specifically in how the model handles medical and wellness queries. GPT-5.3 Instant, the model it replaces, will be retired after a transition period for paid users.

What GPT-5.5 Instant Actually Does Differently

The headline claim from OpenAI is bold: on its most challenging health evaluations, including something called HealthBench Professional, GPT-5.5 Instant now performs on a par with the company’s frontier Thinking models — the slower, more powerful systems previously reserved for higher-stakes tasks.

On the HealthBench benchmark, GPT-5.5 Instant scores 1.8 points higher than its predecessor. On HealthBench Hard, that gap widens to 2.7 points. On HealthBench Professional — the toughest of the set — it improves by 5.5 points. Those aren’t dramatic numbers in isolation, but they represent consistent gains across the board on tests designed to stress-test medical reasoning.

The model is also substantially better at spotting when a situation might need urgent or emergency care. According to OpenAI, it’s more likely to ask users for relevant context before answering, more transparent about uncertainty, and better at translating complex clinical language into something a non-specialist can actually act on.

Fewer Hallucinations, More Caution

One of the most concrete improvements OpenAI is pointing to involves hallucinations — the tendency of AI models to state things confidently that simply aren’t true. On high-stakes prompts covering medicine, law, and finance, GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant, according to internal testing. On difficult conversations previously flagged by users for factual errors, inaccurate claims dropped by 37.3%.

These are OpenAI’s own figures, and they haven’t been independently verified. That caveat matters.

Separately, an OpenAI researcher commented that possible factuality errors detected through privacy-preserving monitoring of health traffic had dropped by around 71% over two months as models improved. Again, this is an internal, unverified figure — but it points in the same direction.

The model is also, perhaps counterintuitively, shorter. GPT-5.5 Instant uses about 30% fewer words and lines on average compared with GPT-5.3 Instant, while OpenAI says it aims to keep the substance and a “warm” conversational tone intact.

Built With Doctors, Not Just Data

To shape the health-specific improvements, OpenAI says it worked with a global network of hundreds of physicians across 60 countries, 49 languages, and 26 medical specialties. Those doctors evaluated where previous model responses fell short — missing red flags, being overconfident, failing to push users towards professional care when they should have.

The result, OpenAI claims, is a model that physicians rated as having fewer failure modes than older versions — and, in some respects, fewer than the physicians themselves. That’s a striking claim, and one that the wider medical and research community is likely to scrutinise carefully.

OpenAI is also explicit about what the model is not. ChatGPT, including GPT-5.5 Instant, is not a substitute for a doctor or emergency services. Its health responses are intended as informational support, not clinical decisions.

Dr Isaac Kohane, chair of biomedical informatics at Harvard Medical School, has previously described the potential for AI in health information as sizeable, while also warning that independent, peer-reviewed validation of these systems is essential before placing significant trust in them. OpenAI’s evaluation process, while involving real physicians, remains largely company-led.

Safety and Biosecurity Flags

GPT-5.5 Instant’s system card — OpenAI’s internal safety documentation — describes the model as “high capability” in biological and chemical safety domains. That means it comes with enhanced guardrails around content that could be misused for harm. OpenAI says it has strengthened those measures as part of this release, though specific details of the safeguards are not publicly exhaustive.

The model also brings broader capability improvements: better handling of image uploads, stronger performance on STEM questions, and tighter web search integration. Those general gains, OpenAI says, underpin the health improvements too.

Reactions and Scepticism

The update is likely to draw a mixed response from clinicians and regulators. GPT-5.5 Instant is a consumer product, not a regulated medical device. It hasn’t been assessed by the Medicines and Healthcare products Regulatory Agency, and NHS England has not endorsed it as part of any clinical pathway.

Privacy advocates will also have questions. Health queries are among the most sensitive data a person can share, and ChatGPT users in the UK are subject to GDPR and UK data protection law. How OpenAI stores, processes, and potentially trains on health conversations remains a point of ongoing scrutiny.

But the scale of use is hard to ignore. If 230 million people a week are already asking ChatGPT health questions — and that figure comes from OpenAI itself — then the quality of those answers matters enormously, whether or not regulators have formally weighed in.

What Happens Next

GPT-5.3 Instant will remain available to paid users for a transition period before being retired. GPT-5.5 Instant is also available as `chat-latest` via OpenAI’s API, meaning developers building health-adjacent applications will pick it up automatically unless they’ve pinned to a specific model version.

OpenAI has signalled that health intelligence will remain a focus area, with further improvements expected as the company continues its physician evaluation programme and internal monitoring.

What This Means for Kent Residents

Anyone in Kent using the free version of ChatGPT is now interacting with GPT-5.5 Instant by default, so if you’ve asked it about a symptom, a medication, or whether something needs urgent attention, you’re getting the updated model. OpenAI’s claimed improvements in recognising when emergency care may be needed could influence whether users contact NHS 111, head to an urgent treatment centre, or attend A&E at hospitals such as Medway Maritime or William Harvey in Ashford — though ChatGPT remains outside regulated NHS clinical pathways, and any AI-generated health information should be checked against NHS guidance or advice from a GP or pharmacist. NHS Kent and Medway ICB has not formally integrated or endorsed ChatGPT as part of local care, so the usual rule applies: if in doubt, call 111 or speak to a clinician.