What OpenAI Dots Do—and Who Can Use Them

OpenAI introduced Dots at DevDay on September 29, 2026, as always-on agents intended to keep working toward a user’s goals between prompts. They run on GPT-6 Astra and can act across apps a person chooses to connect, with the user setting what the agent may do autonomously.23

The launch also raises practical questions: which accounts can use Dots, what their access costs, and how much autonomy to grant an agent working across services. OpenAI’s announcement clarifies some access details, but leaves specific pricing and usage limits unstated.2

How Dots are designed to work

OpenAI says each Dot has its own cloud computer and can work across connected apps. A user gives it a goal and defines the actions it can take on its own. OpenAI also describes teams of Dots working together as a future possibility, not a feature it announced as available now.3

That makes Dots different from an assistant used only for one-off questions: the product is designed for ongoing tasks. But “always-on” describes the intended working model; it does not establish that usage is unlimited or that every task runs without user-defined boundaries.23

Who can access Dots—and what’s known about cost

OpenAI lists Dots for Pro and Business Premium in eligible markets. Enterprise, Education and Healthcare users can try a beta if their workspace administrator enables it; the feature is off by default in those workspaces.2

The DevDay recap names eligible plans and beta conditions but does not give a separate per-Dot price or a specific usage allowance. That means the available announcement supports explaining access eligibility, but not claiming that Dots have a particular standalone price, are included in every subscription, or come with unlimited use.2

Dots run on GPT-6 Astra—not the canceled GPT-6.1 Astra

OpenAI says Dots are powered by GPT-6 Astra.3 The canceled GPT-6.1 Astra was a proposed next-generation upgrade: Reuters reported that OpenAI shelved its release after internal tests found it did not meet the company’s safety and alignment standards.9

These are separate points. The cancellation does not mean Dots run on GPT-6.1 Astra, nor does the reporting establish that Dots themselves failed those tests. It does make clear why permission boundaries matter for agents that can work across connected services: users need to know what they have authorized an agent to do.39

The strategic shift: from answering to acting

Dots extend ChatGPT toward ongoing work across apps, rather than limiting it to responding in a conversation. Reporting has framed the launch as part of a competition with Meta’s Muse for AI agents that can act on users’ behalf.812 The evidence here supports that comparison with Meta; it does not establish a specific Dots-versus-Anthropic product contest.

The launch also came amid broader business growth. Reuters reported that OpenAI’s annualized recurring revenue was approaching $70 billion. That is a reported annualized pace, not revenue already realized over a full year.8

For users and organizations, the practical test is whether an agent can reliably handle useful ongoing tasks while staying within the permissions it has been given. OpenAI’s announcement describes the intended workflow and access tiers, but does not provide a separate price or a detailed usage allowance for Dots.23