ChatGPT's Health Answers Now Beat Doctors', OpenAI Says

OpenAI says its upgraded GPT-5.5 Instant model now writes health answers that outscore ones written by doctors. In the company's own tests, ChatGPT beat physician-written responses across five evaluation categories — accuracy, clarity, and completeness among them — and cut its rate of incorrect health statements by 71% over the past two months.

The model scored 89.9% on an instruction-following benchmark and was measured on HealthBench and HealthBench Professional, OpenAI's own medical evaluation sets. To build and grade it, the company says more than 260 physicians from 60 countries reviewed roughly 700,000 model responses. GPT-5.5 Instant reportedly matches the pricier 'Thinking' models on this work while staying available to free users, with usage limits. OpenAI also rolled out ChatGPT for Clinicians and an OpenAI for Healthcare track — a clear sign it wants a foot inside the clinical workflow, not just the consumer chat.

Why this matters — and the catch

Over 230 million people already ask ChatGPT health questions every week: decoding lab results, prepping for appointments, untangling insurance. So a measurable drop in wrong statements isn't abstract — it reaches a lot of worried people at 2 a.m. The catch is that every number here is OpenAI grading its own homework on its own benchmarks. 'Beats doctors' in a scored eval is not the same as a doctor who can examine you, order tests, and is accountable when they're wrong.

Used well, this is a genuinely useful second reader: great for turning jargon into plain language and for writing down the right questions before you see a real clinician. Used badly, it's a confident voice that talks you out of going at all. The model doesn't know your history, can't touch you, and won't follow up if something changes next week.

I lean on it to translate medicalese before appointments, and it has saved me a few confused phone calls. But I've never once let it replace the appointment itself — and on something as expensive-to-get-wrong as your health, neither should you.