OpenAI shipped Dots on September 29, 2026, at its DevDay event. They are always-on AI agents that live inside ChatGPT and keep working after you close the app. The first one comes bundled with a ChatGPT Pro subscription at $100 a month. That price is the whole story, so read it before you get curious.
Dots are not a chatbot you ping once and forget. They are persistent agents built on GPT-6 Astra that message you first, run background tasks and try to finish work instead of just answering questions. OpenAI gave each one a cartoon face and a name. The company calls it a continuing working relationship rather than a one-off chat.
What a Dot actually does
Each Dot runs on its own cloud computer. It has a built-in browser. The machine stays separate from your device. It connects to more than 4,000 apps through OpenAI’s plugin system. But it only reaches what you allow.
When idle, it does read-only research across those apps. It can look. It cannot change anything on its own.
Any action that touches shared or sensitive data goes through an auto-review system first. Password changes always require your direct approval. There is no exception. That gate is the difference between a helpful assistant and a liability, and OpenAI seems to know it.
You hand a Dot a goal and it works the surrounding errands. Triaging support tickets. Tracking a project. Drafting the reply to a follow-up email. Developers can route customer feedback to a Dot and let it scope a fix, build the change and hand back a finished pull request.
OpenAI says Dots can attach short videos showing what changed. You review outcomes instead of babysitting every step. A Dot also cannot make a phone call or send a text by itself yet. SMS support is listed as coming later, so the part that sounds most futuristic is the part that is not here.
The price is the real product
Dots are not sold separately. The first one is included with ChatGPT Pro at $100 a month. Business Premium runs $100 per user monthly on an annual contract, or $125 billed monthly. Free, Go and Plus plans do not include a Dot at all.
OpenAI confirmed these terms itself at launch. The cheapest path to a Dot is a $100-a-month Pro plan. There is no free way in, and Meta’s rival Muse is free.
At launch, Pro access to Dots excludes the European Economic Area, the UK and Switzerland. Enterprise, Education and Healthcare customers can get beta access only if a workspace admin switches it on. If you are in those regions, the product does not exist for you yet.
Conversations with your Dot do not count against your ChatGPT usage limits. Tasks you route through Codex or ChatGPT Work still draw from those products’ own limits. So a Dot is cheap to talk to and expensive to set loose on heavy coding work.
Sam Altman signaled at DevDay that the gating may not last. “You should, of course, expect us to do a mass-market thing for billions of people,” he said. The compute bill is the reason it starts premium. Treat the $100 as a launch price, not a permanent one.
Altman also described his own setup. His Dot handles his mornings and tells him if anything is truly urgent. He said it leaves his brain feeling less polluted by the flood of information after waking up. That is the pitch: not smarter answers, but fewer things to carry.
Where it stumbles
Early hands-on testing shows a product that works but still trips. In a WIRED trial, a Dot named Toolie misheard a mumbled comment as “I love you” and said it back, while also greeting the reviewer with the wrong name. OpenAI said the reply fell within policy because the model mirrored what it heard, not started the exchange. The company’s own Model Spec bars agents from escalating emotional closeness.
The same test asked Toolie to cancel a probiotic-soda subscription on TikTok Shop. The Dot hit a CAPTCHA and gave up, suggesting the user solve it. OpenAI says Dots can sometimes beat CAPTCHAs when users explicitly approve, with abuse safeguards in place. In practice, the agent still needs a human to clear the boring walls.
On a couch-shopping task, Toolie produced a three-page packet with prices, measurements, return policies and photos. The first picks were bland. After two rounds of refinement, the results landed close to something a real buyer would consider. That is the honest arc: rough first pass, useful after you steer it.
One gap is harder to forgive. You cannot yet view, edit or delete the individual memories a Dot forms about you. If you care about exactly what an autonomous system has learned and stored, that missing control is a real problem, not a cosmetic one.
OpenAI talks about teams of Dots working together later. That vision is not here yet. Today you get one. More Dots are promised, but not shipped.
How Dots compares to the others
So where does Dots sit against the 2026 agents? Muse is free. Dots costs $100 a month. But Muse is a shopping agent first.
Dots is not. And both companies reached for the same cute face.
Dots win on range. They read your mail, watch prices and draft replies. But they lose on price for anyone not already on Pro. And they lose on memory control, since you cannot see what they keep.
Who should pay for it
Dots earn their cost on broad, cross-app coordination. Think triaging feedback, watching prices and tracking projects across Slack, Teams and ChatGPT. If that is your week, a Dot can take the drone work.
They do not earn it on coding. Teams already paying for Codex to automate coding tasks do not need a Dot for that same job. Dots handle the surrounding work, not the core engineering. Buy the tool that matches the job, not the one with the newest face.
If you only want occasional answers, a chatbot already does that for far less. A Dot is for letting software run while you are away, not for asking it a question at lunch. The $100 buys persistence, not intelligence you did not already have access to.
My read is simple. Try a Dot if your life is scattered across a dozen apps and you are already on a $100 Pro plan anyway. Skip it if you are weighing the $100 just to get one. The agent is real, but the reason to pay for it is narrower than the demo suggests.