OpenAI rolls out GPT-5.5 Instant as ChatGPT's default model, reducing hallucinations by 52.5 percent on high-risk topics and improving math reasoning accuracy to 81.2 percent on AIME 2025

OpenAI replaced GPT-5.3 Instant with GPT-5.5 Instant as ChatGPT's default model. In internal testing, GPT-5.5 Instant produced 52.5 percent fewer hallucinated claims than its predecessor on high-risk prompts in medicine, law, and finance, and reduced inaccurate claims by 37.3 percent on tough conversations previously flagged for factual errors.

GPT-5.5 Instant shows measurable gains on competitive benchmarks: AIME 2025 math accuracy jumped from 65.4 to 81.2 percent; GPQA (PhD-level science reasoning) climbed from 78.5 to 85.6 percent; CharXiv (scientific chart reasoning) went from 75.0 to 81.6 percent; MMMU-Pro (expert multimodal questions) rose from 69.2 to 76.0 percent; and OmniDocBench error rate dropped from 14.6 to 12.5 percent. The model also delivers shorter, less verbose answers without losing substance.

A new "memory sources" feature shows users which stored context (saved notes, past chats, or connected Gmail accounts when enabled) shaped a given reply. Users can flag entries as relevant or irrelevant, edit, or delete them. The feature rolls out to all consumer plans on the web first, with mobile to follow.

GPT-5.5 Instant is rolling out to all ChatGPT users immediately. Paying users can access GPT-5.3 Instant through model settings for another three months before it is retired. Enhanced personalization launches first for Plus and Pro users on the web, with Free, Go, Business, and Enterprise plans expected to get access over the coming weeks.