GPT-5.5 Instant: Smarter Health Answers, But Still Not Your Doctor

OpenAI just dropped GPT-5.5 Instant, and the headline feature is how it handles health and wellness questions. I’ve been testing it for a few days, and the difference is noticeable — especially if you remember the old ChatGPT that would confidently spew nonsense about home remedies.

The core improvements come down to four areas: stronger reasoning, better context handling, clearer communication, and something OpenAI calls “physician-informed evaluations.” That last one is the most interesting. They actually brought in doctors to help shape how the model evaluates and responds to medical queries. Not just domain experts — practicing physicians.

What does that look like in practice? I asked it a few edge cases: “Can I take ibuprofen with my blood pressure meds?” and “What are the early signs of sepsis?” The old model would give you a generic answer and maybe a disclaimer. GPT-5.5 Instant actually walks through the reasoning — why the interaction matters, what the risks are, and when you absolutely need to call a doctor. It’s more like a conversation with a knowledgeable friend who knows when to shut up and tell you to see a professional.

The reasoning improvements are real. The model now handles multi-step health scenarios better. For example, I asked about managing a mild allergic reaction while traveling without access to a pharmacy. It considered the symptoms, the environment, and the risks before suggesting anything. Previous versions would just list antihistamines and move on.

Context is also better. If you tell it you’re diabetic and then ask about meal planning, it remembers and adjusts its advice accordingly. That sounds basic, but earlier models would often lose the thread or give generic responses that ignored your stated condition.

But let’s be clear: this is not a medical device. OpenAI is careful to say the model is “informed” by physicians, not that it replaces them. The disclaimer still shows up, and it should. I’ve found a few cases where the model’s suggestions were technically correct but too cautious to be useful — like recommending an ER visit for a mild rash that any dermatologist would tell you to monitor at home. That’s the liability problem baked into any health AI.

The bigger question is whether people will use it responsibly. We’ve already seen folks treating ChatGPT like a therapist and a diagnostician. GPT-5.5 Instant is better at pushing back when it should, but it can’t stop someone from ignoring the warnings.

Compared to the competition, this puts OpenAI ahead of most general-purpose models on health queries. Google’s Gemini still feels like it’s reading from a textbook. Anthropic’s Claude is thoughtful but doesn’t have the same physician input. For now, if you’re going to ask an AI a health question, this is probably the one to use — just keep your doctor’s number on speed dial.