OpenAI has started testing group chats inside ChatGPT, and the pilot hints at a shift in how people might use the tool when more than one person enters the same thread.
The feature sits in a small rollout that touches Japan, New Zealand, South Korea, and Taiwan. It works on the web and on mobile, and it accepts anyone with a logged in account on Free, Go, Plus, or Pro plans. The idea is simple. Put a handful of people in one conversation, let ChatGPT join when asked, and try to keep everything moving in a natural back and forth.
The setup asks for a profile with a name, a username, and a photo before anyone joins a group. Once inside, up to twenty people can talk in the same thread. The tool keeps private chats separate from these shared spaces, so older conversations stay untouched. When someone adds a new participant to an existing chat, the system creates a fresh copy of that thread and leaves the original one alone. The group then appears in a separate section inside the interface so members can find it without scraping through older messages.
OpenAI runs these chats on GPT 5.1 Auto. That means the model chooses the most suitable version based on the prompt and the subscription tier of the person who triggered the response. The AI tries to read the room, so it responds only when the thread needs it and stays quiet when people want to talk among themselves. Mentioning the model by name always pulls it in. It can also react to messages by using emojis and pick up on personal elements like profile photos when users request images that pull in those pictures. The goal seems to be a more social presence inside fast moving threads.
Most of the familiar tools remain active. People can search, upload files, generate images, and speak through dictation. Rate limits hit only when ChatGPT answers. Messages between human members do not count against any usage cap, so the overhead stays light for anyone who just wants to chat with friends or coworkers. Credits come out of the account of whoever triggers the AI, which keeps the cost tied to the person who asked for help.
As per blog post published by Openai, privacy holds a central spot in the pilot. Group chats do not draw on personal memory, and the model does not form new memories inside these shared threads. Everyone can see who is part of the group, and leaving the conversation takes one tap. Only the creator of the group cannot be removed by other participants. Parents can disable group chats for younger users, and anyone under eighteen gets filtered conversations with sensitive content removed for the entire group.
OpenAI’s move lands in a moment when other players are also testing shared or semi shared AI experiences. Microsoft added group chat features to Copilot, and Anthropic introduced ways to share context inside Claude Projects, though that setup does not support real time group conversations. OpenAI’s approach differs because it places several people and the model in one steady thread and keeps the logic of the chat similar to ordinary consumer messaging.
The company frames this pilot as a small first step. It wants to study how people collaborate when an AI sits beside them, not in a separate window. The early design suggests a future where shared plans, research, and debates happen inside one space without switching between apps. OpenAI has hinted that broader access will depend on feedback from these first markets and on how people behave inside the feature. For now, the limited launch gives the company a testbed to watch real world group dynamics and see how well the model handles mixed intentions, sudden shifts, and overlapping voices inside a single conversation.
Note: This post was edited/created using GenAI tools.
Read next: Search Atlas Review: I Tested the AI SEO Platform Powering the Future of Search
The feature sits in a small rollout that touches Japan, New Zealand, South Korea, and Taiwan. It works on the web and on mobile, and it accepts anyone with a logged in account on Free, Go, Plus, or Pro plans. The idea is simple. Put a handful of people in one conversation, let ChatGPT join when asked, and try to keep everything moving in a natural back and forth.
The setup asks for a profile with a name, a username, and a photo before anyone joins a group. Once inside, up to twenty people can talk in the same thread. The tool keeps private chats separate from these shared spaces, so older conversations stay untouched. When someone adds a new participant to an existing chat, the system creates a fresh copy of that thread and leaves the original one alone. The group then appears in a separate section inside the interface so members can find it without scraping through older messages.
OpenAI runs these chats on GPT 5.1 Auto. That means the model chooses the most suitable version based on the prompt and the subscription tier of the person who triggered the response. The AI tries to read the room, so it responds only when the thread needs it and stays quiet when people want to talk among themselves. Mentioning the model by name always pulls it in. It can also react to messages by using emojis and pick up on personal elements like profile photos when users request images that pull in those pictures. The goal seems to be a more social presence inside fast moving threads.
Most of the familiar tools remain active. People can search, upload files, generate images, and speak through dictation. Rate limits hit only when ChatGPT answers. Messages between human members do not count against any usage cap, so the overhead stays light for anyone who just wants to chat with friends or coworkers. Credits come out of the account of whoever triggers the AI, which keeps the cost tied to the person who asked for help.
As per blog post published by Openai, privacy holds a central spot in the pilot. Group chats do not draw on personal memory, and the model does not form new memories inside these shared threads. Everyone can see who is part of the group, and leaving the conversation takes one tap. Only the creator of the group cannot be removed by other participants. Parents can disable group chats for younger users, and anyone under eighteen gets filtered conversations with sensitive content removed for the entire group.
OpenAI’s move lands in a moment when other players are also testing shared or semi shared AI experiences. Microsoft added group chat features to Copilot, and Anthropic introduced ways to share context inside Claude Projects, though that setup does not support real time group conversations. OpenAI’s approach differs because it places several people and the model in one steady thread and keeps the logic of the chat similar to ordinary consumer messaging.
The company frames this pilot as a small first step. It wants to study how people collaborate when an AI sits beside them, not in a separate window. The early design suggests a future where shared plans, research, and debates happen inside one space without switching between apps. OpenAI has hinted that broader access will depend on feedback from these first markets and on how people behave inside the feature. For now, the limited launch gives the company a testbed to watch real world group dynamics and see how well the model handles mixed intentions, sudden shifts, and overlapping voices inside a single conversation.
Note: This post was edited/created using GenAI tools.
Read next: Search Atlas Review: I Tested the AI SEO Platform Powering the Future of Search
