OpenAI has updated ChatGPT’s default model, transitioning from GPT-5.3 Instant to GPT-5.5 Instant for the hundreds of millions of daily users. This change is subtle and devoid of a grand launch or new features but boasts significant enhancements, notably reducing hallucinations by over half.
The GPT family from OpenAI consists of different tiers: Instant for everyday use, Thinking for complex issues, and Pro for high-intensity tasks. GPT-5.5 Instant represents the latest iteration of the most commonly used tier in ChatGPT.
According to OpenAI, GPT-5.5 Instant exhibits fewer hallucinated claims than its predecessor on critical prompts related to medicine, law, and finance—areas where ChatGPT has historically struggled. When evaluated against previously flagged conversations for factual errors, the model saw a 37.3% reduction in inaccuracies.
On HealthBench, which tests AI responses to medical questions with scores ranging from 0 to 100, GPT-5.5 Instant achieved a score of 51.4 points, up from 49.6. In the clinical-use version, HealthBench Professional, it rose from 32.9 to 38.4 points. Given that health inquiries are among the most frequent user requests in ChatGPT, this improvement enhances more than just benchmarks—it increases accuracy, with correct responses now at 38.4%.
Additionally, GPT-5.5 Instant leverages users’ past chats, saved files, and Gmail accounts to generate contextually relevant answers. Users can now see the exact context used and have options to delete or amend it, maintaining control over their data. Temporary chat sessions still allow for complete opt-out.
Previously, Decrypt reported on GPT-5.5’s launch focusing on advanced functionalities like agent-based coding and terminal workflows. However, GPT-5.5 Instant focuses more on everyday tasks such as meal planning and email drafting rather than complex coding pipelines.
The full GPT-5.5 model scores 82.7% on Terminal-Bench 2.0 for command-line task performance, but the Instant version is optimized for general users. Notably, GPT-5.5 Instant is the first in its tier to be classified as “High Capability” in cybersecurity and biological domains, necessitating automated safeguards previously reserved for Thinking models.
GPT-5.3 Instant, which launched in March with promises of reduced refusals and improved accuracy, will remain available to paid subscribers until GPT-5.5 Instant becomes the sole option in three months. Enhanced personalization via Gmail is initially available to Plus and Pro users on the web, with Free, Go, Business, and Enterprise versions rolling out soon.