OpenAI has taken a significant step by making its most medically capable AI model, GPT-5.5 Instant, the default option for all ChatGPT users worldwide. Released on May 5, this model replaces GPT-5.3 Instant and boasts enhanced performance in handling health-related queries, an area where accuracy is crucial.
#How Did GPT-5.5 Instant Improve Medical Accuracy?
The latest model achieved noteworthy scores on OpenAI's internal evaluation tool, HealthBench, where it recorded a range of 49.6 to 51.4 depending on the variant. This reflects a substantial increase of 1.8 points over its predecessor. Particularly for professional-grade medical inquiries, GPT-5.5 Instant outperformed earlier versions by 5.5 points.
The most significant statistic that stands out is the dramatic reduction in hallucinations. OpenAI documented a reduction in fabricated claims by an impressive 52.5% on critical queries related to medical, legal, and financial topics. Furthermore, the number of user-reported factual inaccuracies dropped by 37.3%, indicating that real-world applications of the model are encountering significantly fewer errors.
#What Enhancements Were Made Behind the Scenes?
The broader GPT-5.5 family emerged earlier on April 23, designed with improved reasoning capabilities. The specific variant, GPT-5.5 Instant, was created for rapid responses suitable for everyday discussions while maintaining the sophistication of its predecessor.
For context, OpenAI is implementing a tiered strategy regarding context window sizes. Free users have access to 16K tokens, whereas premium subscribers can utilize up to 128K tokens. This expanded capacity allows users to analyze extensive medical records, research documents, or lengthy legal texts all in a single exchange. Additionally, during the transition phase, paid users can temporarily use the former GPT-5.3 model.