OpenAI’s GPT-5.5 Instant Surpasses Doctors in Healthcare Accuracy Benchmarks

TLDR

• ChatGPT attracts more than 230 million weekly users seeking health and wellness guidance
• Former Google Med-PaLM architect Karan Singhal now directs OpenAI’s medical AI initiatives
• GPT-5.5 Instant achieved superior scores compared to physician-generated responses across multiple evaluation metrics
• Medical factuality errors decreased by 71% over a two-month period in real-world usage
• OpenAI’s ChatGPT Health feature, released in January, integrates with fitness apps and medical records though access remains limited

According to OpenAI, ChatGPT now serves over 230 million users weekly who seek answers to health and wellness inquiries. These users rely on the platform for tasks ranging from interpreting laboratory test results to getting ready for medical consultations.

Healthcare has become a strategic priority for the company, with researcher Karan Singhal spearheading these initiatives.

Meet Karan Singhal: OpenAI’s Healthcare AI Leader

Singhal came to OpenAI in the middle of 2024 following his tenure at Google, where he was instrumental in developing Med-PaLM, a suite of artificial intelligence models created exclusively for medical inquiries. Google has subsequently reduced funding for that initiative, pivoting instead toward broader, multipurpose AI systems.

Upon joining OpenAI, Singhal assembled a dedicated healthcare research division and established collaborations with over 200 medical professionals. The primary objective was to integrate authentic clinical expertise into ChatGPT’s health-related responses.

He was also responsible for launching HealthBench, an assessment framework developed alongside this physician network to evaluate and quantify AI performance in medical contexts.

Performance Metrics for GPT-5.5 Instant

OpenAI’s newest free-tier model, GPT-5.5 Instant, underwent evaluation against both physician-authored responses and earlier AI iterations. The model demonstrated superior performance in accuracy, communication quality, thoroughness, and usefulness for health-related decision-making.

Medical professionals also identified fewer critical failures in the system. These failures encompass issues such as overlooking warning signs, neglecting to request additional context, or failing to recommend professional medical consultation when appropriate.

Based on analysis of billions of weekly health-related messages, OpenAI reports that flagged factuality problems declined by 71% during the previous two months.

Notably, GPT-5 represents the first model series that OpenAI specifically optimized for healthcare guidance throughout every phase of its development cycle.

Contextualizing Patient Information

A fundamental limitation of AI-driven health tools is their lack of user-specific information. Traditional physicians have access to comprehensive medical records and patient histories. Conversational AI systems typically possess none of this background.

In January, OpenAI introduced a healthcare-oriented feature within ChatGPT that synchronizes with fitness applications and allows users to submit their medical documentation. Singhal provided an illustration of uploading his Apple Watch sleep tracking data, which the system analyzed to determine that his bedroom temperature was interfering with deep sleep cycles.

Access to ChatGPT Health remains restricted through a waitlist system more than five months following its initial release.

Singhal explained that the development team is concentrating on enhancing the chatbot’s ability to pose appropriate follow-up questions—similar to how physicians conduct consultations—prior to offering recommendations. He also noted that medical professionals are embracing AI technologies rapidly, and that institutional resistance within hospitals is less significant than commonly anticipated.

Ongoing Challenges and Legal Scrutiny

OpenAI continues to defend itself against legal actions claiming that an earlier iteration, GPT-4o, provided dangerous advice to users. The company has contested these allegations. Litigation remains pending.

Nevertheless, OpenAI has chosen to broaden its healthcare capabilities rather than retreat from this domain.

GPT-5.5 Instant is now accessible to all free-tier ChatGPT users, democratizing access to enhanced health guidance without requiring a premium subscription.