OpenAI Dots: Always-On AI Agents in ChatGPT
OpenAI Dots are always-on agents in ChatGPT with their own computer and browser. What OpenAI confirmed at DevDay, and the four prices it did not.
OpenAI Dots are always-on AI agents that live inside ChatGPT and keep working after you close the window. Announced at DevDay on 29 September 2026, each dot runs on GPT-6 Astra, gets its own cloud computer and browser, and can reach more than 4,000 applications through OpenAI's plugin ecosystem. The first dot is included at no extra charge for Pro and Business Premium customers. If you are thinking about building a workflow on top of them, the more useful fact is what OpenAI left out: the usage allowance, the price of a second dot, and what a faster dot costs.
What a dot actually is
A normal ChatGPT session dies when you leave. A dot does not. OpenAI describes them as agents that keep working after an employee closes the chat window, monitoring projects, using software, reacting to information that changes, and bringing finished work back for approval.
The concrete specification, as announced:
Each dot runs on GPT-6 Astra, not the cheaper Sol line.
Each dot gets its own cloud computer and its own browser, so it is not borrowing your session.
Dots reach more than 4,000 applications through the existing plugin ecosystem, which is what lets one span several tools instead of only reasoning over what is already in the chat.
They appear in the ChatGPT mobile app and website first, with Slack, Microsoft Teams and audio over the phone listed as coming.
The approval step is the design decision worth noticing. A dot is not pitched as something that ships work on your behalf, it is pitched as something that brings work back. That is a meaningfully more conservative posture than the marketing language around autonomous agents usually implies.
Who can turn on OpenAI Dots today
Availability is tiered. Pro and Business Premium customers get their first dot bundled. Enterprise, Edu and Healthcare workspaces can use dots in beta once an administrator enables them. BGR's DevDay roundup lists the eligible plans as Pro 200, Business Premium, Enterprise, Edu and Healthcare.
Dots arrived as one of more than twenty announcements at the event, alongside the GPT-6.1 Sol release and a new premium speed tier. CNBC's live coverage has the running list, including Sam Altman and Sarah Friar taking questions on an eventual IPO.
The four prices OpenAI did not publish
This is the part to write down, because it is the part that decides whether a dot can hold a real job in your business. OpenAI has not disclosed:
The numerical allowance for intensive work, so nobody outside OpenAI knows how much continuous running a bundled dot actually buys.
What an additional dot costs once you want more than the one included.
Pricing for higher speed or higher workload capacity.
Commercial pricing for specialist dots.
An agent whose ceiling you cannot see is fine for experiments and bad for anything with a deadline attached. The failure mode is not a bill you dislike, it is a dot that quietly stops halfway through a task you assumed was covered. Until those numbers exist, treat a bundled dot as a pilot seat, not capacity you can plan a process around.
Where this sits in the agent-coworker race
Dots land in a crowded month. BGR positions them against Meta's Muse, which has a free tier plus monthly plans, and SpaceXAI's Grok Bot. The pattern of vendors selling agents that hold a job over days rather than minutes is not new either: we covered it when Salesforce shipped job-ready agents with a long-horizon runtime, and again when OpenAI started talking about agents working full workdays.
What has changed is packaging. The always-on agent is no longer an API primitive you assemble, it is a seat in a subscription you already pay for. That lowers the barrier to trying one and raises the number of people who will hand one a task without thinking hard about permissions.
What to check before you give a dot real work
Blast radius. A dot with its own browser and 4,000 connectable apps can reach further than the person who configured it intended.
Failure behaviour at the allowance boundary, since that number is unpublished.
Whether the approval step is enforced or merely default, and who can turn it off.
Whether your data-handling commitments to customers survive an agent that browses on your behalf.
If the underlying idea is still fuzzy, our explainer on what agentic AI means in practice covers the mechanics, and how to keep up with AI news is where we keep the method for separating a launch from a shipped feature.
How did this land?
About the author
Senior Editor, AI & Product
Cecilia leads the Swarmz editorial desk. She has spent a decade turning complex AI and product topics into writing people actually finish, and she owns the blog's quality bar.
