OpenAI has entered the personal-agent market with Dots, and the interesting question is not what they can do. It is what the launch does not tell you. The capabilities are on the table; the price of actually using them at scale is not.
What OpenAI Actually Announced
According to MacRumors, Dots are always-on agents inside ChatGPT, announced on September 29, 2026. They run on GPT-6 Astra, can work around the clock, use a cloud computer with a web browser, and connect to more than 4,000 apps. The rollout starts in select markets. Eligible Pro and Business Premium subscribers get one Dot at no extra charge, and Enterprise workspaces get beta access once an administrator turns it on, as TokenPost reports.
The Missing Number
Here is the gap. TokenPost notes that OpenAI has not announced how much workload a bundled Dot includes, nor what extra capacity costs. Users will eventually be able to add Dots and adjust their speed or monthly workload, but the rate card for that dial does not exist yet. For an always-on agent, that is not a footnote: the product is defined by how much it works, so the price of working more is the price of the product.
The Contrast With Meta
Meta’s Muse takes the opposite posture. It is free for most needs, with paid subscriptions only for heavier use, and it is rolling out in the United States on iOS, Android and web. One side bundles an agent into an existing paid plan; the other starts at zero and charges at the margin. Those are two different bets about where the money is: in the subscription you already hold, or in the heavy users who outgrow the free tier.
What To Take From This If You Build or Buy
Do not compare the two on feature lists yet. Both can browse, use apps and run in the background; that is table stakes. The comparison that matters is cost per unit of delegated work, and today only one side has published a clear starting price (free). If you plan to hand a Dot real recurring work, wait for OpenAI to say what the included workload is and what the next increment costs, then measure it against your own tasks.
What this article deliberately leaves out: specific dollar figures for either product’s paid tiers, regional exclusions, and executive quotes about compute. Those appeared in aggregator summaries that contradict each other and could not be confirmed from more than one reliable source, so none are quoted here.
