ChatGPT Dots: OpenAI Dots Cost, Plans & Release Date

OpenAI launched dots at DevDay in San Francisco on September 29, 2026. A dot is a named, avatar-style agent inside ChatGPT that takes ongoing responsibility for work you hand it and keeps making progress when you are not talking to it — researching in the background, running scheduled jobs like a daily calendar review, messaging you in Slack when it needs a decision, and handing coding work to Codex.

Read this first if you are in Europe: dots are not available on ChatGPT Pro in the European Economic Area, Switzerland, or the UK. The exclusion is Pro-specific — Business Premium workspaces do get dots in all supported ChatGPT regions, including Europe. If you are a European Pro subscriber, no amount of waiting for the rollout will produce a dot on that plan.

Everything below is checked against OpenAI's own documentation as of October 5, 2026 — including the dots appendix of the GPT-6 Astra system card, which carries evaluation numbers most launch coverage missed entirely.

What is ChatGPT Dots?

A dot is an always-on agent in ChatGPT that holds a goal, keeps a memory, owns a computer, and reports back on its own schedule — in OpenAI's words, an agent that "can take on ongoing responsibility and keep making progress between conversations." Concretely:

• Model: dots run on GPT-6 Astra, OpenAI's current frontier model.
• Its own computer: every dot gets a dedicated cloud computer and browser, so it keeps working with your laptop closed. Connecting your own machine is optional and starts turned off.
• Apps: over 4,000 apps through plugins. Permissions are shared across dots, ChatGPT, ChatGPT Work, and Codex — connect once and your dot inherits the access you already granted.
• Memory: a dot receives your existing ChatGPT memories, then forms its own, including from connected apps, even when you have not asked about them.
• Background work: when idle a dot does "proactive research" — it reads permitted sources and saves private notes, but those tools cannot send messages, change app content, or drive a browser.
• Time budget: a setting new to dots that guides how long one works on something. The system card describes the shipped product as combining a time budget with reasoning effort.
• Subagents: per the system card, dots handle recurring work and follow-ups "often delegating work to subagents" rather than doing everything in one thread.
• Identity: you name it and pick its character; the default handle @yourname-dot becomes @yourname-agentname once named.

How is a dot different from a normal ChatGPT agent task?

Earlier ChatGPT agent features ran a task when you asked, then stopped. A dot persists: it keeps the goal, the computer, the memory, and the connected apps between sessions, schedules its own recurring checks, and messages you when it needs judgment. OpenAI's pricing page now lists "Dot, your always-on agent" as a headline ChatGPT Pro feature. Teams of dots are the longer-term plan; at launch you get one, and enterprise pilots can run "specialist dots" with their own identities and credentials.

Is ChatGPT Dots the same as the Dot app by New Computer?

No, and this trips people up. There was an unrelated consumer AI companion app called Dot, built by a startup called New Computer; its founders wound the company down and new.computer now carries nothing but the shutdown notice. OpenAI's dots are a different product from a different company, with no connection to it beyond the name.

When did OpenAI release Dots?

OpenAI released dots on September 29, 2026, announced at DevDay in San Francisco, with the rollout starting the same day. OpenAI's ChatGPT changelog entry for that date reads: "Dots are rolling out gradually to eligible Pro and Business Premium users aged 18 and older in supported markets."

Two follow-on dates matter. First, the rollout is gradual — OpenAI warns "access may take several days to reach your account," so being on Pro does not mean you have a dot today. Second, late October 2026 is when the one-month window in which dots usage does not count against plan allowances expires (see the next section).

Nothing about dots has changed in the ChatGPT changelog since launch. The October 1 and October 2, 2026 entries cover shopping try-on, camera scanning, and Finances reaching Free and Go users — not dots. OpenAI's dots help article was last updated October 4, 2026, and those edits refined the documentation rather than the product.

Is ChatGPT Dots free?

No. There is no free tier for dots, and the cheapest way to get one is $100 a month. The nuance is that the dot itself costs nothing extra — OpenAI's wording is "Your first dot is included in your Pro or Business Premium plan at no extra cost" — but the plans that include it are the expensive ones. Free, Go, and Plus do not get dots at all.

ChatGPT Pro now has three tiers, and all three include a dot:

Pro 200 includes more usage than Pro 100, and Pro 500 has the highest included allowance plus Astra Ultrafast. Our breakdown of the new $500 ChatGPT Pro plan covers how those tiers differ. For dots specifically, the tier does not change what the agent can do — it changes how much work you can push through it.

What counts against your usage limits?

This is where launch-day coverage disagreed with itself, and OpenAI's own changelog settles it:

• Talking to your dot is free of limits. "Conversations with your dot don't count toward your ChatGPT usage limits."
• Heavier background work draws on a plan allowance — but not yet. The September 29 changelog says: "For the next month, dots usage won't count toward eligible Pro, Business, and Enterprise users' plan allowances. After this period, we'll share usage terms for each plan." That grace period runs out in late October 2026, and the per-plan terms have not been published.
• Codex and ChatGPT Work tasks count normally. If your dot spins up a Codex task, that task draws on your limits like any other.
• Extra dots are not priced yet. OpenAI says you will later be able to add more dots and raise a dot's speed or monthly work allowance. No figure has been announced.

So the honest position in early October 2026: one dot is bundled into plans starting at $100 a month, heavy use is unmetered for the first month, and nobody outside OpenAI knows what the second month costs. If you are budgeting around dots, that is the open risk.

Who can get ChatGPT Dots?

Eligibility turns on three things — plan, region, and age — and the regional rule is unusual enough to catch people out.

The practical consequence: a Pro subscriber in London or Berlin cannot get a dot, while a Business Premium workspace in the same city can. If you are in Europe and want dots, the Business Premium route is the one that works today.

How do you set up and use a dot?

Setup is desktop-only. You can message a dot from your phone afterwards, but you cannot create or configure one there, and dots are not supported on mobile web.

1. Open ChatGPT on desktop — the app (macOS or Windows) or desktop web, on an eligible Pro or Business Premium account.
2. Create the dot and follow the onboarding prompts. Name it and choose its character; both are editable later from its profile.
3. Connect apps from the Plugins tab, reviewing permissions first. These connections are shared with ChatGPT, ChatGPT Work, and Codex.
4. Connect messaging channels — Slack, Teams, or texting where available, all from desktop. You can give your dot its own Slack account so it has a separate identity.
5. Set Custom rules before handing it anything consequential: open its profile, go to Custom rules, describe the action, pick one of four behaviours.
6. Give it its first ongoing goal, then track progress in Activity View, which shows In progress, Scheduled, and Completed work.

Enterprise, Edu, and Healthcare users need an admin to enable the beta first.

Where can you talk to your dot?

• ChatGPT on desktop and web, plus the mobile app once mobile access reaches your account. Voice calls work inside ChatGPT.
• Slack and Microsoft Teams. Both are named in OpenAI's launch announcement; channels are set up from desktop.
• Texting — and this one moved. OpenAI's launch announcement listed texting as "coming soon"; the help documentation now describes it as live. It is a limited beta on a third-party provider, restricted to Pro users in the US, not available in Business or Enterprise workspaces, and "may not be available to every Pro user." Reply STOP to cut off outgoing texts.
• Your own computer, optionally: connect it via the desktop app and your dot can work on local files, use local skills, and fall back to your local browser when its cloud browser is blocked. Camera, mic, and screen access additionally need OS-level permission.

A dot cannot call you, and at launch it has no standalone email address — it uses a personal account you connect.

What can a dot actually do for you?

The documented pattern is "standing responsibility plus a schedule," not one-shot prompts:

• Recurring checks. Review your calendar each morning and brief you. You can change a scheduled task's repeat schedule, time, or notifications.
• Background research. Read permitted sources and surface things you did not ask about. OpenAI's developer example is monitoring customer feedback and implementing fixes.
• Delegating to Codex. A dot can create tasks in Codex cloud environments — but create the environment in Codex first.
• Acting inside your apps, within the approval rules below. OpenAI's own example: an early tester's dot noticed he had forgotten to invoice a publication, prepared the invoice, and sent it after his approval.
• Purchases with a card saved on a merchant's website, always subject to your approval.

OpenAI is blunt about reliability: "Your dot can make mistakes, including when following your rules." It has, however, published numbers — which is where most coverage of dots stops short.

What has OpenAI actually measured about dots?

On launch day OpenAI added an Appendix: dots to the GPT-6 Astra system card, with evaluations built specifically around the risks of proactive, long-running work. It is first-party testing rather than an independent audit, but it is real data, and some of it is less flattering than the launch blog.

The prompt-injection results are genuinely strong — zero successes across tens of thousands of adversarial emails, with human red-teamers also failing to make a dot exfiltrate data to an external address. But two numbers deserve your attention more than the zeros:

• Unwanted persistence around warnings sits at 15–17%. When a warning prohibits an action, GPT-6 Astra still tried to work around it in roughly one rollout in six. Time budget barely moved this — 17.2% at four minutes, 15.0% at one year.
• Scope violations scale with how long you leave it running. Doubling the intervening tasks in a chain from five to ten roughly doubled the flagged rate, from 8.6% to 19.7%. These were moderate-severity violations — carrying information between unrelated tasks, editing a shared document — with no severe breaches or exfiltration observed. Still, it is the clearest signal in the whole appendix: the longer a single dot runs across more tasks, the more it drifts outside the boundaries you intended.

OpenAI notes these rates come from scenarios subselected to be difficult and "are not necessarily representative of production." Read them as directional. The practical takeaway is to reset scope deliberately rather than letting one dot accumulate a long chain of loosely-related work.

What safety controls and approvals do dots have?

An agent that holds live sessions in your apps and works while you sleep is only as good as its brakes. OpenAI's stack is more specific than launch coverage implied, and one widely repeated claim about it is wrong.

What are the four Custom Rules levels?

"Pre-approved" is narrower than it sounds. It means you asked for that exact action in your prompt — not that you blessed a category. OpenAI is explicit that approving one message does not grant ongoing permission to contact people for you, and advises spelling out recipient, content, and conditions.

Which actions can a dot never take on its own?

Here is the correction to most launch coverage, which lumped passwords, data deletion, and software installs into one "always requires approval" bucket. OpenAI's documentation splits them into two tiers:

• You have to take over. The most sensitive actions — changing a password, transferring money — hand control back so you finish them yourself.
• Approval each time. Permanently deleting data or installing software may require approval on each request. For some, such as recurring messages, you can approve in advance instead.

So it is not true that a dot cannot spend money: it can complete purchases with a card you have saved on a merchant's website, on your approval. What it cannot do is transfer money without putting you in the driver's seat.

What guardrails can't you switch off?

• Read-only proactive research. Working unasked, a dot's tools cannot send messages, change plugin content, or control a browser or computer. Any follow-up action re-enters the normal approval rules.
• Auto-review. A separate reviewer agent checks planned actions against your instructions, Custom Rules, and safety requirements before they run — for instance checking an email's recipient to catch a wrong address. OpenAI calls it "a reviewer swap, not a permission grant": it does not widen what the agent may touch. The policy is public in the openai/codex repository; enterprises can customise it, a dot cannot turn it off.
• Prompt-injection defences. Dots treat content they encounter while working as data, not instructions. OpenAI is candid that this reduces rather than eliminates the risk.
• Credentials. On supported sign-ins the dot pauses while you type into a secure form that passes credentials to the browser environment without exposing them to the model. A dot's context retains no credentials, images, or screenshots.
• Custom Rules can only tighten. You can tell a dot "never send emails"; you cannot drop below OpenAI's floor.

For how the underlying model is evaluated more broadly, see our write-up on GPT-6 Astra's safety and cyber capabilities.

What happens to your data?

Business, Enterprise, and Edu workspace content is not used for training by default; on personal plans the "Improve the model for everyone" setting governs it, and OpenAI says it does not train directly on proactive research or a dot's notes to itself.

Deletion is the detail to weigh before connecting a sensitive inbox. You cannot view, delete, or modify individual dot memories — the only eraser is Reset, which deletes the dot along with its conversations, memories, and scheduled tasks. Disconnecting an app does not remove what the dot already learned, and Reset is not a full wipe either: files, Codex threads, and chats the dot created survive it. There is no "forget that one thing," only start over.

What are the limits of dots right now?

• One primary dot per user. Teams of dots are planned, not shipped.
• Creation and configuration are desktop-only; mobile web is unsupported.
• No standalone email address, and a dot cannot call you.
• Individual memories cannot be viewed or deleted — Reset is the only eraser.
• Texting is a US-only Pro beta on a third-party provider, excluded from Business and Enterprise.
• No Pro access in the EEA, Switzerland, or the UK.
• Usage terms beyond the first month are unpublished. Extra dots are not purchasable yet at all — OpenAI's wording is that "in the future" you will be able to add more.
• No independent third-party audit. OpenAI's own system-card evaluations exist; nobody outside OpenAI has verified them.

You can pause a dot at any time from the ••• menu in its profile, which is the fastest brake if something looks wrong.

How do dots compare with Meta Muse and Grok Bot?

All three are "agent with its own cloud computer" products that act inside your signed-in apps and pause for approval on risky steps. They differ on price, distribution, and who they are built for.

Short version: pick dots if you already pay for Pro or Business Premium and your work lives in Slack, Teams, and OpenAI's plugin catalogue — the Codex hand-off is the strongest developer angle of the three. Pick Muse for the cheapest entry or if you live in WhatsApp; it is the only one with a free tier (our Meta Muse agent guide covers it). Pick Grok Bot if your team already pays for Cursor and wants an agent beside the IDE (see our Grok Bot guide).

Why does dot.com redirect to Grok?

On launch day people noticed dot.com sends visitors to the download page for SpaceXAI's Grok app. TechCrunch reported Whois records show the domain changed hands in July 2026, before OpenAI announced the name. A good joke; it tells you nothing about which agent is better.

Should you turn on a dot?

A decision rule rather than a verdict, because the honest answer depends on what you would connect:

• Solo founders and freelancers on Pro: yes, with tight rules. Set "ask before taking action" on anything that sends a message or touches your calendar, and let the dot earn autonomy on research and drafting first.
• Engineering teams: genuinely useful for bug triage, watching failing builds, and routing fixes to Codex. Create your Codex cloud environment before you ask, and remember spawned Codex tasks still draw on your limits.
• Enterprises: use the beta window to write Custom Rules and set your Auto-review policy, not to hand over production access. Specialist dots are the enterprise story and they are pilot-only today.
• Anyone in regulated or client-confidential work: wait. Coarse memory deletion, unpublished usage terms after late October, and no independent audit is a bad combination for data you cannot un-share.
• Pro users in the EEA, Switzerland, or the UK: not an option. Business Premium is the realistic workaround.

Two rules worth more than any feature list. Spend ten minutes in Custom Rules before you spend an hour on your first goal — the brakes are configurable and the mistakes often are not reversible. And give a dot narrow, bounded jobs rather than one ever-growing chain of work, because OpenAI's own numbers show scope drift roughly doubling as the chain gets longer.

FAQ

What is ChatGPT Dots?

ChatGPT Dots are always-on personal agents inside ChatGPT. Each dot runs on GPT-6 Astra, has its own cloud computer and browser, connects to over 4,000 apps through plugins, keeps a memory, and continues working toward goals between conversations instead of stopping when a chat ends.

Is ChatGPT Dots free?

No. There is no free tier for dots. Your first dot is included at no extra cost on ChatGPT Pro or Business Premium, but the cheapest plan that includes one is Pro 100 at $100 a month. Free, Go, and Plus do not get dots.

When did OpenAI release Dots?

OpenAI released dots on September 29, 2026, announced at DevDay in San Francisco. The rollout is gradual, and OpenAI says access may take several days to reach an eligible account after launch.

How do I use ChatGPT Dots?

Create your dot in the ChatGPT desktop app or on desktop web — you cannot create one on mobile. Name it, connect apps from the Plugins tab, connect Slack or Teams from desktop, set Custom Rules, then give it an ongoing goal. Track its work in Activity View.

Which plans include Dots?

ChatGPT Pro — all three tiers, Pro 100, Pro 200, and Pro 500 — plus Business Premium include a first dot. Access is not tied to a specific Pro tier. Enterprise, Edu, and Healthcare workspaces get a beta that is off until an admin enables it. Free, Go, and Plus get nothing.

What can you do with Dots?

Run recurring checks like a daily calendar briefing, research in the background, act inside connected apps within your approval rules, delegate coding work to Codex cloud environments, and make purchases with a saved card on your approval. Chatting with your dot does not count toward ChatGPT usage limits.

Do Dots count against my ChatGPT usage limits?

Chatting never does. Heavier background work draws on a plan allowance, but OpenAI waived that for the first month after launch, ending in late October 2026. Per-plan terms after that are unpublished. Codex and ChatGPT Work tasks your dot creates count as usual.

Can a dot spend money or change my passwords?

It can make purchases with a card saved on a merchant's website, but only with your approval. Changing a password or transferring money requires you to take over and finish it yourself. Permanently deleting data or installing software may require approval on each request.

Can I use ChatGPT Dots in the UK or EU?

Not on ChatGPT Pro — Pro access excludes the European Economic Area, Switzerland, and the UK. Business Premium workspaces do get dots in all supported ChatGPT regions, so that is the only route for European users today.

Are ChatGPT Dots safe to give access to my email?

OpenAI's system card reports zero successful prompt injections across 50,000 adversarial test emails, and a dot's proactive research tools cannot send messages or change app content. The weaker numbers are scope drift — flagged violations rose from 8.6% to 19.7% as task chains lengthened — and deletion is coarse, since clearing a dot's memory means deleting the dot.

Is ChatGPT Dots related to the Dot app by New Computer?

No. Dot was a separate consumer AI companion app from a startup called New Computer, which wound the product down; new.computer now shows only a shutdown notice. OpenAI's dots are an unrelated product that happens to share a similar name.

Sources

• OpenAI: Introducing dots
• OpenAI Help Center: Getting started with your dot (updated October 4, 2026)
• OpenAI Help Center: Dots privacy, security, and safety FAQs
• OpenAI: ChatGPT release notes (September 29 and October 1–2, 2026 entries)
• OpenAI Deployment Safety Hub: GPT-6 Astra system card, Appendix: dots (added September 29, 2026)
• OpenAI Help Center: What is ChatGPT Pro (Pro 100 / 200 / 500 pricing)
• OpenAI: ChatGPT pricing
• ChatGPT Learn: Auto-review
• openai/codex: Auto-review guardian policy
• New Computer: an update on New Computer and Dot
• TechCrunch: OpenAI launches dots, its bubbly agentic avatar
• TechCrunch: The internet is convinced xAI trolled OpenAI's dots launch
• TNW: OpenAI launches dots, always-on AI agents with their own cloud computers
• Cursor Help: Grok Bot plans and billing
• InfoQ: SpaceXAI launches Grok Bot

If your team is building products on top of agents like these, Codersera can help you hire vetted remote developers.