Author: Louise Ann "LA" Maca

OpenAI just updated ChatGPT with GPT-5.5 Instant, and the company says the new model gave healthcare answers that scored higher than doctors’ answers in OpenAI’s own tests for accuracy, clarity, and completeness.

ChatGPT Download

OpenAI also says the error rate in medical statements has fallen by 71% over the past two months. It says GPT-5.5 Instant is better at picking up on cases where you may need urgent care, asking for missing context, and admitting uncertainty instead of sounding more confident than it should.

That matters if the scale is anywhere near what OpenAI reports: 230 million people use ChatGPT every week for health-related questions. In a small independent check using 18 medical prompts, GPT-5.5 Instant produced 54.5% fewer hallucinated claims than GPT-5.3 Instant.

OpenAI says those results fit a broader review process in which more than 260 physicians from 60 countries evaluated over 700,000 responses.

If you use ChatGPT for health questions, this is an update you’ll probably want to look at. Still, be careful. Internal benchmarks can make OpenAI’s model look stronger than it really is, and fewer hallucinations don’t erase the risk of dangerous mistakes. Under-triage, delayed visits to a clinician, privacy issues, and bias are all still real concerns.

You can use GPT-5.5 Instant in ChatGPT now, but OpenAI is still presenting it as a tool to help you make sense of symptoms, questions, or treatment options before talking to a qualified medical professional, not as a replacement for medical care.

Google just picked up an important court win in Berlin over Google AI Overviews. The judges said these AI-generated summaries count as a search-results format, not as content Google itself has published.

The Berlin court said Google doesn’t have “decisive influence” over the exact wording, and that users can recognize the Overviews as compiled summaries built from third-party sources. That weakened a perfume maker’s claim that its brands were showing up alongside cheaper knockoffs and links to competing sellers.

Berlin also breaks from what happened in Munich. There, two publishers were wrongly tied to scams and fraudulent business practices. The Munich court found that the summaries contained independent, new, and substantive statements, and it issued a temporary injunction.

So Germany now has a real split on this question: are these outputs just aggregation, or are they generation? If you follow AI search, this one matters. Any wider move toward treating these answers as Google’s own statements would leave the company more exposed when the summaries are false, misleading, or commercially damaging.

The Munich decision isn’t final, and Google is expected to appeal. It’s also already dealing with wider pressure in Europe, including an antitrust complaint from the European Publishers Council and a separate copyright dispute connected to Hungary over the reuse of journalistic content. For now, though, the Berlin ruling gives the company some breathing room.