OpenAI introduced Dots last Tuesday, September 29, as its direct answer to Meta's Muse agent. Unlike previous assistants — which responded only when invoked — Dots are "always-on" agents running on dedicated virtual machines in the cloud, powered by GPT-6 Astra, the company's most expensive and capable model. That model, priced at $10 per million input tokens and $50 per million output tokens on the API, is the first broadly available GPT-6 generation, succeeding GPT-5.6 Sol and introducing computer use and browser navigation with a 1.05 million token context window.
But what draws the most attention about Dots is not the underlying technology, but the access model: users pay for the base plan, not the agent itself. The cheapest eligible tier is ChatGPT Pro 100 at $100 per month, which includes one primary dot at no additional cost. Higher tiers like Pro 200 ($200/month) and Pro 500 ($500/month, newly introduced with Ultrafast speed) also qualify, as does Business Premium and Enterprise. Free, Go, and Plus plans receive no Dots at launch. The most visible geographic restriction covers the European Economic Area, Switzerland, and the United Kingdom — likely due to data privacy concerns that Meta also faces with Muse.
Each dot operates on an isolated virtual machine with its own browser, can connect to more than 4,000 applications through OpenAI's plugin ecosystem, and learns user preferences over time — a function that inevitably results in collecting enormous amounts of personal data. OpenAI states that conversations with the dot do not count against ChatGPT usage limits, but deeper tasks run on a "quota" whose exact limit has not yet been revealed. For one month after launch, neither conversations nor tasks count against the quota — but post-evaluation terms remain open, raising the question of how much these agents will cost under sustained use.
The Dots architecture traces its roots to OpenClaw, an open-source project that OpenAI acquired through acqui-hire and turned into a product. The difference is that while OpenClaw ran on the user's local machine, Dots run on infrastructure controlled by OpenAI. Each dot has built-in safety rules: when working in the background, it can only read connected apps — it cannot send messages, alter content, or control the user's browser or computer. More sensitive actions, like changing passwords, remain with the user. The company is also working with Microsoft to integrate "specialist dots" with Microsoft Agent 365 security controls, which could position the product as a direct competitor to established enterprise tools.
The competition is fierce. Meta's Muse is already available for free and reached more than 30 million active users within weeks, despite privacy controversies. xAI's Grok Bot costs just $20 per month with Cursor Pro — five times less than Dots. There are also smaller agents like Tab, which exited stealth with a $300 million valuation. Dots' minimum price positions OpenAI as the premium product in the sector, which may work well for businesses already invested in ChatGPT Business, but creates a significant barrier for individual consumers who have not yet seen concrete proof of ongoing value.
What Dots reveal about the future of personal AI is that the race for always-on agents is no longer about responsiveness — it's about persistence, memory, and context. Each dot remembers past conversations, previous decisions, and the user's working style. This means that instead of an assistant that answers questions, we now have an assistant that "works" even when we are not watching. It is the same reasoning that led Apple to develop the Neural Engine for on-device computing, but applied at datacenter scale with a model capable of executing thousands of operations per month on dedicated servers.
The real monthly cost of running Dots continuously remains uncertain. If the "deep work" quota is low, users may find that each additional task costs much more than the advertised base price. If OpenAI does not publish clear limits by the end of the evaluation period (October 29 for existing Pro 200 users, dates not yet announced for Pro 100), cost ambiguity could generate frustration similar to what Meta faces with Muse. And there is the data question: if Dots learn user preferences over time and personal data is collected for this learning, how does OpenAI ensure this information will not be used to improve its models unintentionally? The Dots privacy control system allows users to choose whether conversations are used for training — but the default policy remains a blind spot. Whether $100 a month justifies the investment when free or cheap competition is evolving so rapidly remains an open question.
Sources: The Verge, The New York Times, Fortune
✓ Independent sources cross-checked and verified before publishing
