GPT-5.5 Instant Scores Higher Than Physicians in OpenAI Health Accuracy Tests

TLDR

• Over 230 million people use ChatGPT for health and wellness advice every week
• Researcher Karan Singhal leads OpenAI’s healthcare push, previously built Google’s Med-PaLM
• GPT-5.5 Instant outperformed physician-written answers in accuracy and communication tests
• Factuality issues in health responses dropped 71% in two months based on production traffic
• ChatGPT Health, launched in January, connects to health apps and medical records but still has a waitlist

OpenAI says more than 230 million people now use ChatGPT for health and wellness questions every week. That includes everything from reading lab results to preparing for doctor appointments.

The company has made healthcare a core focus, and researcher Karan Singhal is leading that effort.

Who Is Karan Singhal?

Singhal joined OpenAI in mid-2024 after working at Google, where he helped build Med-PaLM, a set of AI models designed specifically for medical questions. Google has since scaled back investment in that project, shifting toward general-purpose models.

At OpenAI, Singhal built a new health research team and partnered with more than 200 physicians. The goal was to bring real medical judgment into how ChatGPT responds to health questions.

He helped launch HealthBench, an evaluation tool built with that physician group to test and measure AI health performance.

How GPT-5.5 Instant Performs

OpenAI’s latest free model, GPT-5.5 Instant, was tested against physician-written answers and older models. It scored higher across accuracy, communication, completeness, and health decision helpfulness.

Physicians also rated it as having fewer failure modes. That includes things like missing red flags, not asking for more context, or failing to refer users to care when needed.

OpenAI said that based on billions of health messages a week, the rate of flagged factuality issues dropped 71% over the last two months.

GPT-5 is also the first model family OpenAI trained at every stage of development specifically to improve health advice.

Knowing More About the Patient

One of the challenges with AI health tools is that they know very little about the user. A doctor has your records and history. A chatbot usually has none of that.

In January, OpenAI launched a health-focused feature inside ChatGPT that connects to health apps and lets users upload medical records. Singhal gave an example of uploading his Apple Watch sleep data, which the tool used to identify that his bedroom temperature was disrupting deep sleep.

ChatGPT Health still has a waitlist more than five months after launch.

Singhal said the team is focused on making the chatbot ask better questions, the way a doctor would, before giving advice. He also said doctors are adopting AI tools quickly, and that hospital resistance to AI is lower than many expect.

Pressure to Keep Improving

OpenAI is still facing lawsuits alleging that an older model, GPT-4o, gave harmful advice to users. The company has denied liability. Those cases are ongoing.

Despite that, OpenAI has continued to expand health features rather than pull back.

GPT-5.5 Instant is available to all free ChatGPT users, making improved health responses accessible without a paid subscription.

The post GPT-5.5 Instant Scores Higher Than Physicians in OpenAI Health Accuracy Tests appeared first on CoinCentral.