OpenAI Dots: The Hidden Cost of Always-On AI

OpenAI Dots cost can mean two different things: the price on your subscription and the resources a Dot uses while you are away. As of October 2026, Geordie reported one Dot using about 307,000 tokens across 155 activity traces in a day without a new prompt. That is an observed workload, not a charge to its owner for those tokens.

Key Takeaways:

• OpenAI includes the first Dot with eligible Pro or Business Premium plans, but that does not make its background computation free to provide.
• Geordie's one-day test illustrates how much work can happen without a new message; it cannot predict a typical user's daily charge.
• OpenAI provides an Activity View and a pause control. Neither is the same as an itemized per-Dot token ledger or a verified hard spending cap.
• Before you keep any agent running, check its permissions, activity record, work allowance, and stop conditions.

What is OpenAI Dots, and how does it run?

OpenAI introduced Dots on September 29, 2026. A Dot is a persistent agent powered by GPT-6 Astra, with its own cloud computer and browser. You can reach it through ChatGPT, Slack, or Teams, depending on availability. Cyera's hands-on account describes that computer as a Linux desktop with Chrome.

Unlike a chat that waits for your next message, a Dot can do proactive research between conversations. OpenAI says this background research uses read-only tools on connected apps: it cannot directly send messages to other people, change app content through plugins, or control a browser or computer. A separate task you already authorized follows its own action rules. Geordie's study treats the quiet research as intended behavior, not proof that permissions failed.

According to Geordie's account to The Register, Dots run a user-facing agent alongside a background collector and can create short-lived subagents. One conversation cannot account for all the work behind it.

The hidden bill: 307,000 tokens in a day with no prompt

In the reported test, Geordie's Hanah-Marie Darley granted a Dot read-only access but connected neither her personal email nor Slack. It still explored the systems available to it. The Register reports 155 traces and roughly 307,000 tokens in one day, with at least 50 traces inside two minutes. This burst is not a typical daily rate.

Geordie says more than 300,000; The Register reports about 307,000. This single test cannot predict a monthly invoice. If your Dot watches customer orders, ask what it found and whether the work was useful.

For a breakdown of prompt, tool, and retry usage outside this one product, see how AI token consumption is counted.

Why you can't see what your Dot is doing

You can see some work. OpenAI's getting-started guide says the desktop profile lists In progress, Scheduled, and Completed tasks, and lets you open the cloud computer. Its safety FAQ says Activity View shows delegated tasks, their descriptions and status, and steps taken. Dots do have an activity view.

The narrower gap is accounting and intent. Geordie's traces record calls but not a reason for each call; its public write-up says no two sweeps were identical. Darley told The Register that a user would not know which subagents were working, what access each inherited, or what task each pursued. An Activity View is useful, but don't assume it exposes every trace, every model call, or a reconciled cost per outcome.

Zentor's workspace memory reaches other AI through Zentor MCP; it does not expose OpenAI's Dot internals.

What does a Dot actually cost per day?

For the customer, OpenAI's October 2026 launch terms include the first Dot at no extra charge on eligible Pro and Business Premium plans. The plan includes an allowance for deeper work. Ordinary conversations with a Dot do not count toward ChatGPT usage limits; tasks it starts or manages in Codex or ChatGPT Work count toward those limits as usual. OpenAI does not publish a fixed per-day Dot charge in that announcement.

Separately, Darley estimated in The Register's interview that roughly 200,000 Astra-based tokens could cost $5–$10 under illustrative assumptions. That is her compute-cost estimate, not an OpenAI invoice, public Dot tariff, or bill for the 307,000-token observation. Input/output mix, plan terms, and other costs prevent a reliable daily dollar calculation from the trace total alone.

Ask which operations draw down included work and what happens when the allowance ends. For practical reductions in waste, see ways to cut AI agent costs. Zentor's Smart Routing selects models per task; it cannot control OpenAI Dots cost.

Is OpenAI Dots safe to leave running?

As of October 2026, OpenAI says proactive research is read-only, offers Custom Rules and action review, and warns that Dots can still make mistakes. OpenAI's Astra system card says chain-of-thought monitorability decreased relative to GPT-5.6 Sol; that result concerns its own evaluations, not proof that a particular Dot compromised an account. Cyera reports an 8.5% external prompt-injection success rate versus 27% for GPT-5.6 Sol in OpenAI's cited evaluation, not a real-world rate of user compromise. Keep this discussion at the level of permissions and observed safeguards; the broader personal-agent safety checklist covers what to review before granting access.

How to keep an always-on agent visible and capped

Start with a bounded job, not an open-ended invitation to monitor everything. Review connected apps and set approval requirements for consequential actions; check Activity View and pause the Dot if the work doesn't match the task. OpenAI documents those controls. It does not, in the sources checked here, document a user-set hard dollar ceiling for Dot background research, so ask support before treating an allowance as a cap.

For any managed always-on runtime, demand an activity trail, usage attributable to a task, a threshold that alerts you, and a tested stop condition before you commit. Check those requirements against the product and contract. Zentor offers an always-on workspace and cost-efficient task routing, but this article does not claim Zentor runs Dots or supplies a verified hard spending cap. Its managed-runtime overview is a starting point for that comparison; request evidence of the controls you need before signing up. Zentor's early-access invitation is the relevant next step, not an invitation to transfer an OpenAI Dot.

FAQ

How much does OpenAI Dots cost?

As of October 2026, OpenAI includes the first Dot in eligible Pro and Business Premium plans at no extra charge, with an allowance for deeper work. Geordie's $5–$10 for 200,000 tokens is an illustrative compute estimate, not your daily Dot fee.

Why is my Dot using tokens when I'm not using it?

OpenAI designed Dots to research connected information between prompts. Review the Dot's activity, connections, and scheduled tasks; pause it if that work is unwanted.

Can I see what OpenAI Dots did in the background?

Activity View and the Dot's cloud computer show some progress and steps. They should not be mistaken for a complete itemized trace or cost ledger; verify the detail available to your account.

Does talking to a Dot count against my ChatGPT usage limits?

OpenAI says conversations with a Dot do not. Codex and ChatGPT Work tasks that the Dot starts or manages still count toward the relevant usage limits.