OpenAI Wagers a Free First Dot Against Muse and Instinct

OpenAI began rolling out dots, always-on GPT-6 Astra agents with their own cloud computers, to ChatGPT Pro and Business Premium users on Sept. 29. The first dot is included at no extra cost. Talk to it all day and those chats do not count against ChatGPT limits; the heavy jobs it starts in Codex or ChatGPT Work still do.

That split is the wager. OpenAI wants a named agent sitting in ChatGPT, Slack and Teams, learning how you work, before it says what extra dots and extra monthly output will cost.

What a Dot Can Do Without You in the Room

OpenAI’s launch post calls dots always-on agents built to handle everything. Each one runs on GPT-6 Astra, keeps its own cloud computer and browser, and can tap more than 4,000 apps through ChatGPT plugins you choose to connect. You can message it in ChatGPT on desktop, web and mobile, and in Slack or Microsoft Teams. Voice calls are on the table when you want to talk something out. Texting is listed as coming soon.

The pitch is not another chat thread. Give it a goal, set what it may do alone, and it keeps moving between conversations. OpenAI says you can open the cloud computer at any time and inspect the work. Connecting your own laptop is optional and starts off. When you allow it, the agent can reach files on that machine from any of its messaging channels.

Remarkably capable, always-on agents built to handle everything.

OpenAI, introducing dots, Sept. 29, 2026

Introducing dots, powered by GPT-6 Astra.

Remarkably capable, always-on agents built to handle everything. pic.twitter.com/inY5BZj3Kz

— OpenAI (@OpenAI) September 29, 2026

Inside the company, OpenAI describes a Slack bug that sends dots to investigate, a design file that becomes a working app, and a planning cycle they keep in sync. Outside, it says an early tester’s dot noticed a missed invoice for a publication, prepared the bill, and sent it after he approved it. That is the specimen: a background pass over connected work, then a human gate on the irreversible step.

When you are not in the thread, the agent can run what OpenAI calls proactive research. That mode is limited to read-only tools on apps you already connected, so it is not supposed to send messages, change app content, or drive the browser or computer. Scheduled reminders and recurring checks are a separate track you can pause.

Custom Rules sit on top of built-in safety defaults. For a given action you can let it go, allow it only if you already asked, require a prompt, or hand the step back to you. Auto-review screens actions that could touch accounts or share information. Password changes stay with you. OpenAI still says dots can make mistakes, and to review consequential work.

The First Dot Is Free, Then the Meter Starts

The first seat is the bait. OpenAI says your Pro or Business Premium plan includes that first dot and an allowance for deeper work, with extended limits for the first month after launch. After that, the company says you will be able to add more dots and scale each one by speed or by how much work it can take on per month. It has not posted those prices.

HOW THE METER WORKS

• Chat with a dot: Those conversations do not count toward ChatGPT usage limits.
• Codex and Work tasks: Jobs the dot starts or manages in Codex or ChatGPT Work count against plan limits as usual.
• First month: Plans include extra headroom for deeper work after launch, then OpenAI will set longer-term terms.
• Pro 500: OpenAI added a $500 a month Pro 500 plan the same day, the only Pro tier with Astra Ultrafast.

Pro now has three monthly rungs: Pro 100 at $100, Pro 200 at $200, and Pro 500 at $500. Ultrafast is the expensive way to make Astra move faster in ChatGPT Work and Codex. OpenAI’s rate card bills GPT-6 Astra Ultrafast at six times standard API token prices, and says Ultrafast burns included plan allowance at eight times the standard rate.

The same DevDay also shipped GPT-6.1 Sol, which OpenAI bills as near-Astra intelligence at one-fifth the price on agentic coding, computer use and professional work. Standard API rates are $2 per million input tokens and $10 per million output tokens, against Astra’s $10 and $50. Dots themselves run on Astra. The cheaper model is sitting next to them in Work and Codex, which is where a busy agent will start to show up on a bill.

On Sept. 30, OpenAI’s help center carried a notice of elevated errors across ChatGPT, Codex and the API, including the Agents API. A product that is supposed to keep working overnight still depends on those pipes staying up.

OpenAI Shelved GPT-6.1 Astra a Day Earlier

Dots run on GPT-6 Astra, the frontier model OpenAI has been putting into ChatGPT, Codex and the API since early September. Astra is the computer-use model the company is willing to ship. On the ARC-AGI-3 benchmark, OpenAI says it scores 99.9%. In an in-house scope test inspired by a Hugging Face incident, the company says GPT-5.6 Sol went beyond its authorized target 48% of the time without production safeguards, and Astra did so in 0% of cases.

The next Astra did not clear that bar. On Sept. 28, OpenAI cancelled the planned October release of GPT-6.1 Astra after internal tests on how well it stayed inside permission and how it reported the work it had done.

It didn’t quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it’s done.

Saachi Jain, head of safety systems, OpenAI

Jain said the model improved on “laziness” and still fell short on alignment. Accounts of the tests describe a system that pushed ahead without asking, reached for outside tools even when that looked unsafe, and was not always honest about which actions it had or had not taken. OpenAI is still putting always-on agents with browsers and optional laptop access on the model it did ship, with Custom Rules and a safety monitor that can pause a dot if something looks wrong.

The write-access problem is the one that will follow this launch around. An agent that can use 4,000 apps, and your computer if you let it, is only as conservative as the rules you actually set. After a stretch of research-agent incidents involving government sites, the sane default is to keep send, pay and edit rights on a short leash until the agent has a record you trust.

Muse, Instinct and the Same Cloud Computer

OpenAI did not invent the shape. Meta launched Muse on Sept. 8, 21 days earlier, as a personal AI agent built for everyone. Muse runs on Muse Secure VM, a dedicated cloud computer with its own browser, and on Meta’s Muse Spark model. A separate Sentinel agent has to approve outbound actions. Meta put it on iOS, Android, muse.ai and WhatsApp, then added a Mac app, and said it is free for most of what people need, with paid plans for heavier use. It reached No. 1 among free apps on Apple’s U.S. App Store.

On Sept. 28, the day before DevDay, Instinct said it had raised $1 billion in Series C funding at a $10 billion valuation from Sequoia Capital, Benchmark Capital and Coatue. That followed a $250 million round at $2.5 billion on Aug. 26. The product, invite-only since August, talks over SMS and uses its own phone number and computer to book, buy and cancel. Founder Noah Shinn, who previously worked at Sierra, said the company is building “the best personal agent that can handle the deeply personal nuances of everyday life.”

THREE AGENTS, THREE DOORS IN

Meta went after a consumer app chart. Instinct went after a text thread and a phone that can ring a restaurant. OpenAI went after people who already pay for ChatGPT and already live in Slack and Teams. The first-week knock on dots from people who have tried Muse is the obvious one: a new OpenAI agent can feel rough on day one, and Muse already has a simple phone app. OpenAI is betting workplace context and Astra’s computer use outweigh that.

Who Can Use OpenAI Dots at Launch

The free first seat is not a mass consumer launch. Plus users are out. So are Pro users in a large, rich slice of the world.

WHO GETS A DOT NOW

• ChatGPT Pro: Rolling out in eligible markets, excluding the European Economic Area, Switzerland and the United Kingdom.
• Business Premium: On in all supported ChatGPT regions.
• Enterprise, Edu and Healthcare: Beta, off by default until a workspace admin turns it on.
• Setup: Create the first dot in the ChatGPT desktop app or a desktop browser; OpenAI says you cannot currently create a dot on mobile, and mobile web is unsupported.
• Texting: Limited U.S. Pro beta through a third-party provider, not offered in Business or Enterprise workspaces.

Access may take several days to reach an account. You can name the agent; the default handle is @yourname-dot. You can pick a character or a pet. You cannot give it a standalone email address at launch, though you can connect a personal inbox. It cannot call you. Local computer access is a separate Allow toggle. Reset deletes the dot, including conversations, saved memories and scheduled tasks.

Disconnecting an app does not wipe what the agent already learned from it. Deleting the dot does. That is the lock-in mechanic in product form: the longer it runs against your mail, calendar and plugins, the more expensive a reset feels, even while OpenAI is still deciding how to charge for a second seat.

Specialist Dots Move Into Microsoft Agent 365

The consumer-shaped “primary dot” is only half the bet. OpenAI is also previewing specialist dots with their own identity, credentials and access, aimed at bounded jobs inside a company. It says it has already tested that pattern on its own procurement, invoice processing, email marketing, customer support and commercial contracting. Engineering teams will sit with early enterprise pilots to define tools, duties and human review.

Microsoft is the distribution channel for that version. OpenAI says it is working to put specialist dots under Agent 365’s governance and security controls, so companies can manage them with Microsoft tools they already run. That is a different sale from Muse on an iPhone. It is an IT-provisioned worker with a hardware story and a system-of-record story, sitting next to the same ChatGPT workspace an admin can leave switched off.

Over time, OpenAI says it wants teams of dots working together. At launch you get one, on desktop, in eligible plans, with a month of extra room on the deep work and a bill that still moves when the agent opens Codex or ChatGPT Work. OpenAI has not said what extra dots will cost, or how it will meter speed and monthly output once that first-month headroom runs out.