OpenAI Makes GPT-5.5 Instant ChatGPT's New Default: 52.5% Fewer Hallucinations

GPT-5.5 Instant Replaces GPT-5.3 Instant as Default

OpenAI rolled out GPT-5.5 Instant as ChatGPT's new default model on May 5, replacing GPT-5.3 Instant. The update applies to all ChatGPT users and is also available through the API.

52.5% Fewer Hallucinations on High-Stakes Prompts

According to OpenAI, GPT-5.5 Instant produces 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts in medicine, law, and finance. On difficult conversations that users had previously flagged for factual errors, inaccurate claims dropped by 37.3%.

Memory Sources and Gmail Personalization

Plus and Pro users now receive responses personalized using past conversations, uploaded files, and Gmail, accessed through ChatGPT's search tool. A new "memory sources" control, rolling out across all models, shows which context was used to generate a given answer — and lets users delete or correct outdated sources. Mobile support is planned for shortly after the web rollout.

Broader Performance Gains

The upgrade improves visual reasoning, math, science, STEM queries, uploaded image analysis, and web-search decision-making. For casual advice prompts, GPT-5.5 Instant reduces unnecessary follow-up questions and heavy formatting, producing tighter responses.

Source: TechCrunch · Axios