OpenAI's GPT-5.5 Instant improves health answers in ChatGPT

OpenAI has upgraded the default artificial intelligence model powering its ChatGPT service, significantly improving its performance on medical and health-related queries. The new model, GPT-5.5 Instant, is now available to all users, both free and paid, and the company reports it matches the accuracy of its most advanced "frontier" reasoning models on key health benchmarks while dramatically reducing the rate of fabricated or incorrect information.

The upgrade, which became the default for ChatGPT on Thursday, June 18, 2026, represents a major step in addressing one of the most critical areas for AI chatbots: providing reliable medical information. OpenAI states that on its internal medical evaluation suite, HealthBench, GPT-5.5 Instant scored between 49.6 and 51.4, showing improvement over its predecessor. More notably, the company recorded a 52.5% decrease in "hallucinated" claims-confidently stated falsehoods-on high-stakes prompts covering medical, legal, and financial topics. User-reported factual errors in live traffic also fell by 37.3%.

In a more direct comparison, OpenAI had a panel of physicians evaluate responses from the new model against responses written by other...