Particle.news
Download on the App Store

OpenAI Tests ChatGPT Group Chats in Four Countries, Powered by GPT-5.1

The trial brings the assistant into shared conversations of up to 20 people under privacy safeguards.

Overview

  • Group chats are live on web and mobile for users in Japan, South Korea, Taiwan and New Zealand, allowing up to 20 participants to message with ChatGPT in a single thread.
  • ChatGPT in groups is tuned to behave more socially, speaking only when mentioned or when it detects a useful task such as summarizing, organizing lists or offering options, and it runs on GPT-5.1 Auto.
  • Privacy measures include memory being disabled in groups, additional protections for under‑18 users and the option for guardians to block group use, with personal chat histories kept separate.
  • Alongside the pilot, OpenAI is rolling out GPT-5.1 to paid tiers first with Instant, Thinking and Auto variants, new API endpoints (gpt-5.1 and gpt-5.1-chat-latest) and a plan to retire GPT-5 over the next three months.
  • OpenAI reports higher accuracy from GPT-5.1, including 94.6% on AIME 2025 math without tools, stronger programming scores, about 45% fewer factual errors with web search and roughly 80% lower error rates in extended reasoning versus the o3 model.