ChatGPT Dots Review 2026: OpenAI's Always-On Agent

How it scores

What we love (and don't)

OpenAI launched Dots on September 29, 2026. The announcement was short on hype and long on specificity: Dots are “remarkably capable, always-on agents built to handle everything.” Each dot runs on GPT-6 Astra, gets its own cloud computer, and keeps working between conversations.

That last part is what matters. Not a chat session that extends until you close the tab. A persistent agent that continues toward a goal after you walk away.

The question worth asking is not whether Dots is impressive. It is. The question is what kind of user has paid $100 a month, what work actually benefits from persistent delegation, and whether the current limits are tolerable enough to build a workflow around.

What Dots actually does

Each dot has its own cloud computer and its own browser. It connects to apps you authorize through OpenAI’s plugin ecosystem, which covers more than 4,000 apps. Once connected, it can reach those apps to gather information, complete tasks, and make progress without you watching.

You can reach your dot through ChatGPT on desktop, web, and mobile, through Slack, and through Teams. The dot carries context across all of them.

OpenAI separates the time you spend working with your dot from the time you don’t. When you’re away, the dot does what OpenAI calls “proactive research,” using your connected apps through read-only tools. It can’t send messages, change app content, or control your browser or computer. When it does act, built-in rules decide when it acts on its own and when it asks for your approval.

That boundary is the part I would point a cautious buyer to. In the background, a dot can only read. When it acts, auto-review checks anything that could affect your accounts or share information, and Custom Rules let you tighten what it may do without asking.

Plans and pricing

Dots are not available on Free, Go, or Plus. On the ChatGPT pricing page, the dot shows up on the Pro plan card and on Business Premium seats, and nowhere on the Free, Go, or Plus cards.

Pro starts at $100 per month. The plan offers three usage tiers, and $100 is the floor. Your first dot is included.

Business Premium seats cost $100 per month billed annually, or $125 per month if billed monthly. That compares to Business Standard seats at $20 per month (annual) or $25 per month (monthly). The premium is $80 per seat per month on an annual plan. The main additions are five times more usage, no five-hour usage limit, and access to a dot.

Business Standard users don’t get a dot. This is worth being precise about. The Business plan has two seat types. Only the more expensive one includes Dots.

Enterprise pricing is custom. Enterprise users, including Edu and Healthcare, can try the Dots beta when their workspace admin enables it.

The limits question

OpenAI is clear on one thing and quiet on another.

Clear: your first dot is included in Pro or Business Premium at no extra cost, and the plan includes an allowance for deeper work with extended limits for the first month after launch.

Quiet: what the limits look like after that first month. OpenAI’s announcement says you’ll eventually be able to add more dots and scale the output of each one, but it gives no numbers for limits after the first month.

One more distinction: conversations with your dot don’t count toward your ChatGPT usage limits. But if you ask your dot to kick off tasks in Codex or ChatGPT Work, those count against your existing plan usage. The dot conversation is free. The downstream work is not.

Security and privacy

The cloud computer isolation is the most defensible security claim. Your dot runs on its own remote machine. Your local computer and its contents stay separate unless you explicitly connect them. For supported site logins, the dot can use saved passwords without exposing them to the model. The credential stays outside the model’s context.

For data training: Business, Enterprise, and Edu workspaces are not used to train OpenAI’s models by default. On personal plans, you control whether your dot’s conversations and work are used to improve the models. OpenAI is also explicit that it does not train directly on a dot’s proactive research or internal notes.

The monitoring layer is worth noting. OpenAI’s security safeguards defend against malicious instructions and monitor for harmful behavior. If the monitoring system flags something, it can pause or stop the dot’s work.

Custom Rules let you go further: you can specify which actions require your approval, which are allowed automatically, and which are blocked entirely.

Who Dots is for

The user Dots is designed for has a persistent, delegatable workload and no patience for re-briefing an AI every session. A developer who wants a persistent agent monitoring feedback and prepping pull requests. A product manager who wants launch materials revised as the product scope changes, without directing each update. A researcher who wants analysis updated as new data arrives.

None of those are edge cases. They’re real workflows. The question is whether you have enough of them to justify the price point and whether you’re on Pro or Business Premium already.

If you’re on Plus at $20 per month, Dots requires upgrading to Pro at a minimum of $100 per month. That’s an $80 monthly delta. The justification needs to come from somewhere in your actual work, not from the feature sounding good.

For teams evaluating Business plans, the seat math is more tractable. Moving from Standard ($20/month annual) to Premium ($100/month annual) costs $80 per seat per month more. For users whose work involves enough multi-step delegation to benefit from persistent agents, that math can work. For users doing straightforward chat and document work, Standard is probably adequate.

Alternatives

To me, the always-on agent category still feels new. Most of the AI tools I review wait for a prompt. Dots is built to keep going without one.

Other large AI companies are building in the same direction. Meta introduced Muse, its personal AI agent, on September 8, 2026, and says its new Muse Wonder model will soon power it. SpaceXAI’s X Subscription, announced October 12, bundles Grok, Grok Bot, Cursor, and X Premium into one plan. Our October 14 roundup covers both.

For context on the GPT-6 Astra model that powers Dots, our GPT-6 Astra analysis covers the model and the benchmark claims around it. That article covers the model side; this one covers the agentic deployment.

If Dots interests you but ChatGPT itself has been giving you trouble, the ChatGPT troubleshooting guide covers the most common errors.

If you’re on a lower plan and looking for basic task automation inside ChatGPT, scheduled tasks (listed on the Plus plan, and expanded on Pro) are a more accessible starting point. Not persistent, not agentic, but cheaper and less complex.

The honest read

Dots is a real product making a real claim. The claim is that it can keep work moving while you’re not watching, under rules you control. The architecture behind that claim is coherent: isolated cloud computer, read-only background mode, monitoring with pause capability.

The catch is the access gate. At $100 a month for the lowest qualifying plan, Dots is not for most ChatGPT users. The plan cards are explicit: Free, Go, and Plus do not list a dot.

That’s not a criticism. It’s a positioning choice. OpenAI knows who this is for. They priced it accordingly.

If your work fits the profile and your plan qualifies, Dots is worth setting up on day one. If you’re on Plus or lower, the upgrade math needs to do real work in your budget before this makes sense.

FAQ